IJACSA Volume 15 Issue 8 - thesai.org

IJACSA Volume 15 Issue 8

Copyright Statement: This is an open access publication licensed under a Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, even commercially as long as the original work is properly cited.

View Full Issue

Paper 1: Ensemble Learning with Sleep Mode Management to Enhance Anomaly Detection in IoT Environment

Abstract: The rapid proliferation of Internet of Things (IoT) devices has underscored the critical need for energy-efficient cybersecurity measures. This presents the dual challenge of maintaining robust security while minimizing power consumption. Thus, this paper proposes enhancing the machine learning performance through Ensemble Techniques with Sleep Mode Management (ELSM) approach for IoT Intrusion Detection Systems (IDS). The main challenge lies in the high-power consumption attributed to continuous monitoring in traditional IDS setups. ELSM addresses this challenge by introducing a sophisticated sleep-awake mechanism, activating the IDS system only during anomaly detection events, effectively minimizing energy expenditure during periods of normal network operation. By strategically managing the sleep modes of IoT devices, ELSM significantly conserves energy without compromising security vigilance. Moreover, achieving high detection accuracy with limited computational resources poses another problem in IoT security. To overcome this challenge, ELSM employs ensemble learning techniques with a novel voting mechanism. This mechanism integrates the outputs of six different anomaly detection algorithms, using their collective intelligence to enhance prediction accuracy and overall system performance. By combining the strengths of multiple algorithms, ELSM adapts dynamically to evolving threat landscapes and diverse IoT environments. The efficacy of the proposed ELSM model is rigorously evaluated using the IoT Botnets Attack Detection Dataset, a benchmark dataset representing real-world IoT security scenarios, where it achieves an impressive 99.97% accuracy in detecting intrusions while efficiently managing power consumption.

Author 1: Khawlah Harahsheh

Author 2: Rami Al-Naimat

Author 3: Malek Alzaqebah

Author 4: Salam Shreem

Author 5: Esraa Aldreabi

Author 6: Chung-Hao Chen

Keywords: IoT; IDS; machine learning; ensemble technique; sleep-awake cycle; cybersecurity; anomaly detection

Paper 2: Optimizing Low-Resource Zero-Shot Event Argument Classification with Flash-Attention and Global Constraints Enhanced ALBERT Model

Abstract: Event Argument Classification (EAC) is an essential subtask of event extraction. Most previous supervised models rely on costly annotations, and reducing the demand for computa-tional and data resources in resource-constrained environments is a significant challenge within the field. We propose a Zero-Shot EAC model, ALBERT-F, which leverages the efficiency of the ALBERT architecture combined with the Flash-Attention mechanism. This novel integration aims to address the limita-tions of traditional EAC methods, which often require extensive manual annotations and significant computational resources. The ALBERT-F model simplifies the design by factorizing embedding parameters, while Flash-Attention enhances computational speed and reduces memory access overhead. With the addition of global constraints and prompting, ALBERT-F improves the generaliz-ability of the model to unseen events. Our experiments on the ACE dataset show that ALBERT-F outperforms the Zero-shot BERT baseline by achieving at least a 3.4% increase in F1 score. Moreover, the model demonstrates a substantial reduction in GPU memory consumption by 75.1% and processing time by 33.3%, underscoring its suitability for environments with constrained resources.

Author 1: Tongyue Sun

Author 2: Jiayi Xiao

Keywords: Artificial intelligence; natural language processing; event argument classification; zero-shot learning; flash-Attention; global constraints; low-resource

Paper 3: A Personalized Hybrid Tourist Destination Recommendation System: An Integration of Emotion and Sentiment Approach

Abstract: This research introduces a personalized hybrid tourist destination recommendation system tailored for the growing trend of independent travel, which leverages social media data for trip planning. The system sets itself apart from traditional models by incorporating both emotional and sentiment data from social platforms to create customized travel experiences. The proposed approach utilizes Machine Learning techniques to improve recommendation accuracy, employing Collaborative Filtering for emotional pattern recognition and Content-based Filtering for sentiment-driven destination analysis. This integration results in a sophisticated weighted hybrid model that effectively balances the strengths of both filtering techniques. Empirical evaluations produced RMSE, MAE, and MSE scores of 0.301, 0.317, and 0.311, respectively, indicating the system's superior performance in predicting user preferences and interpreting emotional data. These findings highlight a significant advancement over previous recommendation systems, demonstrating how the integration of emotional and sentiment analysis can not only improve accuracy but also enhance user satisfaction by providing more personalized and contextually relevant travel suggestions. Furthermore, this study underscores the broader implications of such analysis in various industries, opening new avenues for future research and practical implementation in fields where personalized recommendations are crucial for enhancing user experience and engagement.

Author 1: Suphitcha Chanrueang

Author 2: Sotarat Thammaboosadee

Author 3: Hongnian Yu

Keywords: Recommendations; hybrid recommendation system; Collaborative Filtering; Content-based Filtering; social media data; travel planning

Paper 4: Automatic Identification and Evaluation of Rural Landscape Features Based on U-net

Abstract: The study delves into the landscape feature identification method and its application in Xijingyu Village, investigating landscape composition elements. Analyzing rural landscape structure holistically aids in dividing landscape characteristic zoning maps, essential for guiding rural landscape and territorial spatial planning. By utilizing GIS software for superposition analysis based on topography, geology, vegetation cover, and land use, the village range of west well valley undergoes further refinement. To address the inefficiencies of common foreground extraction algorithms relying heavily on rural landscape images, a novel approach is introduced. This new algorithm focuses on directly extracting foreground areas from rural landscape interference images by leveraging stripe sinusoidal characteristics. An adaptive gray scale mask is established to capture the sinusoidal changes in interference stripes, facilitating the direct extraction of foreground areas through a calculated blend of masks. In evaluating the results, the newly proposed algorithm demonstrates significant improvements in operation efficiency while maintaining accuracy. Specific enhancements include classifying pixel gray values into intervals and recalibrating them to enhance analysis metrics. Compared to traditional methods, the algorithm showcases advantageous enhancements across various parameters, such as PRI, GCE, and VOI. Moreover, to address challenges in unwrapping low-quality rural landscape phase areas, a ResU-net convolutional neural network is employed for phase unwrapping. By constructing image datasets of interference stripe wrapping and unwrapping alongside noise simulations for model training, the network structure's feasibility is verified. The study's innovative methodologies aim to optimize rural landscape analysis and planning processes by enhancing accuracy and efficiency in landscape feature identification, foreground area extraction, and phase unwrapping of rural landscapes. These advancements offer substantial improvements in quality and precision for territorial spatial planning and rural landscape management practices.

Author 1: Ling Sun

Author 2: Jun Liu

Author 3: Yi Qu

Author 4: Jiashun Jiang

Author 5: Bin Huang

Keywords: Rural landscape; foreground area extraction; deep learning; phase unwrapping; ResU-net

Paper 5: Analysis of Customer Behavior Characteristics and Optimization of Online Advertising Based on Deep Reinforcement Learning

Abstract: With the shift from traditional media to online advertising, real-time strategies have become crucial, evolving to meet contemporary demands. Advertisers strive to succeed in online advertising evaluations by demand-side platforms to secure display opportunities. Discrepancies in information evaluation can impact click-through rates, emphasizing the need for precise prediction models in asymmetric contexts. Time dynamics significantly influence online ad click-through rates, with rest hours outperforming working hours. This study introduces the ARMA model to refine click predictions by preprocessing hits and employing a single XGBoost model. Furthermore, a reinforcement learning model is developed to explore online advertising strategies amidst information imbalances. Data is segmented into training (70%), validation (15%), and test sets (15%), with model parameters optimized using the DQN algorithm over 48 hours. Validation and testing on separate datasets comprising 15,000 entries each yield model accuracies of 0.85 and recall rates of 0.82. The incorporation of regret minimization algorithms enhances reward functions in deep reinforcement learning. Leveraging Tencent data, a comparative analysis evaluates advertisers’ click rates as overrated, underrated, or accurately predicted by DSPs. Findings indicate that smart customer behavior characteristics outperform DQN, converging swiftly to optimal solutions under complete information. Smart characteristics exhibit stability and flexibility, with human-machine collaboration circumventing the drawbacks of random exploration. Transfer Learning amalgamates experimentation with real-world insights, bolstering algorithm adaptability for intelligent decision-making tools in enterprises.

Author 1: Zhenyan Shang

Author 2: Bi Ge

Keywords: Real-time online advertising; ARMA-XGBoost model; information asymmetry; deep reinforcement learning decision-making behavior; Transfer Learning

Paper 6: Logistics Transportation Vehicle Monitoring and Scheduling Based on the Internet of Things and Cloud Computing

Abstract: This paper addresses challenges in the logistics industry, particularly information lag, inefficient resource allocation, and poor management, exacerbated by global economic integration and e-commerce growth. An advanced logistics and transportation vehicle monitoring and scheduling system is designed using IoT and cloud computing technologies. This system integrates Yolov5 for real-time vehicle location, DeepSort for continuous tracking, and a space-time convolutional network for vehicle status analysis, forming a comprehensive monitoring model. An improved multi-objective particle swarm optimization algorithm optimizes vehicle scheduling, balancing objectives like minimizing travel distance, time, and carbon emissions. Experimental results demonstrate superior performance in real-time monitoring accuracy, scheduling efficiency, arrival time prediction, road condition forecasting, and failure risk prediction. Notable achievements include 95% vehicle utilization, a 0.25 RMSE for predicted arrival times, and a 0.20 MAE for failure risk prediction. While the system significantly enhances operational efficiency and supports resource optimization, future work will focus on data security, system stability, and practical deployment challenges. This research contributes to transforming the logistics industry into a smarter, greener, and more efficient sector.

Author 1: Kang Wang

Author 2: Xin Wang

Keywords: Internet of Things; cloud computing; logistics and transportation; vehicle monitoring; vehicle scheduling

Paper 7: Performance and Accuracy Research of the Large Language Models

Abstract: Starting with the end of 2022, there has been a massive global interest in Artificial Intelligence and, in particular, in the technology of large language models. These reduced the resolution of many problems dailies of varying degrees of complexity at a level accessible to every individual, whether it was an academic, business or social environment. A multitude of digital products have begun to use large language models to offer new functionalities such as intelligent messaging applications trained to respond efficiently depending on the specific parameters of a company, virtual assistants for programmers (GitHub Copilot), video call summarization functionality (Zoom), interpretation and extraction rapid drawing of conclusions from massive data (Big Data). These are just a few of the many uses of these technologies. Therefore, the general objective of this paper is the comparative analysis between three large language models such as ChatGPT, Gemini, and Llama3. Each model's strengths and constraints are analyzed, offering insights into their optimal use cases. This analysis provides a comprehensive understanding of the current state of large language models powered by deep learning, capable of executing various natural language processing (NLP) tasks, guiding future developments and applications in the field of artificial intelligence (AI).

Author 1: Nicoleta Cristina GAITAN

Keywords: Large language models; artificial intelligence; ChatGPT; natural language processing

Paper 8: Image Generation Using StyleVGG19-NST Generative Adversarial Networks

Abstract: Creating new image styles from the content of existing images is challenging to conventional Generative Adversarial Networks (GANs), due to their inability to generate high-quality image resolutions. The study aims to create top-notch images that seamlessly blend the style of one image with another without losing its style to artefacts. This research integrates Style Generative Adversarial Networks with Visual Geometry Group 19 (VGG19) and Neural Style Transfer (NST) to address this challenging issue. The styleGAN is employed to generate high-quality images, the VGG19 model is used to extract features from the image and NST is used for style transfer. Experiments were conducted on curated COCO masks and publicly available CelebFace art image datasets. The outcomes of the proposed approach when contrasted with alternative simulation techniques, indicated that the CelebFace dataset results produced an Inception Score (IS) of 16.57, Frecher Inception Distance (FID) of 18.33, Peak Signal-to-Noise Ratio (PSNR) of 28.33, Structural Similarity Index Measure (SSIM) of 0.93. While the curated dataset yields high IS scores of 11.67, low FID scores of 21.49, PSNR of 29.98, and SSIM of 0.98. This result indicates that artists can generate a variety of artistic styles with less effort without losing the key features of artefacts with the proposed method.

Author 1: Dorcas Oladayo Esan

Author 2: Pius Adewale Owolawi

Author 3: Chunling Tu

Keywords: Artworks; VGG19; Neural Style Transfer; Generative Adversarial Network; inception score; StyleGAN

Paper 9: Applied to Art and Design Scene Visual Comprehension and Recognition Algorithm Research

Abstract: Combining advanced intelligent algorithms to improve the scene visual understanding and recognition method for art design can not only provide more inspirations and creative materials for artists, but also improve the efficiency and quality of art creation, and provide scientific and accurate references of artworks. Focusing on the art design scene visual understanding and recognition problem, a scene visual understanding and recognition method based on the intelligent optimisation algorithm to optimise the structural parameters of the multilayer perception machine is proposed. Firstly, the scene visual recognition method is outlined and analyzed, and the application scheme of multilayer perceptron in the understanding and recognition problem is designed; then, for the problems of the multilayer perceptron model, such as the training does not generalize, combined with the Pond's optimization algorithm, the training parameters of the multilayer perceptron model are optimized, and the visual understanding and recognition scheme of the art design scene is designed; finally, the proposed model is verified with the image dataset, and the scene visual understanding and recognition accuracy reaches 0.98, compared with other models, the proposed method has higher recognition accuracy. This research solves the problem of scene visual understanding and recognition, and applies it to the field of art design to improve the efficiency of art design assistance.

Author 1: Yuxin Shi

Keywords: Art design; scene visual understanding and recognition; multilayer perceptron; pond goose algorithm; image dataset

Paper 10: Quantitative Measurement and Preference Research of Urban Landscape Environmental Image Based on Computer Vision

Abstract: At present, research on landscape preferences mostly uses traditional questionnaire surveys to obtain public aesthetic attitudes, and the analysis method still relies on manual coding with small sample sizes. However, the research on landscape preference of applying network big data and computer vision technology is rare, and the research content and algorithm application are limited. In order to improve the research effect of quantitative measurement and preference of urban landscape environment image, the algorithm proposed in this paper combines two-dimensional analysis modules, two-dimensional visual domain analysis and three-dimensional visual analysis, and makes full use of the advantages of the two analysis modules, and analyzes the scale from large scale to medium and micro scale based on different accuracy urban digital models. Through image classification and content recognition, image semantic segmentation and image color quantification, the landscape feature information in pictures is mined, and the dimension of landscape image is put forward based on this. In addition, this paper combines experimental analysis to verify that the method proposed in this paper has certain results. It is not only suitable for visual analysis of landmark buildings and landmark structures in cities, but also can analyze the visual characteristics of natural landscapes as urban images in cities. Therefore, the quantitative method of urban visual landscape analysis proposed in this paper can provide reliable data support for the follow-up urban design work.

Author 1: Yan Wang

Keywords: Computer vision; urban landscape; environmental image; quantification; measure

Paper 11: Clustering Algorithms to Analyse Smart City Traffic Data

Abstract: Urban transportation systems encounter significant challenges in extracting meaningful traffic patterns from extensive historical datasets, a critical aspect of smart city initiatives. This paper addresses the challenge of analyzing and understanding these patterns by employing various clustering techniques on hourly urban traffic flow data. The principal aim is to develop a model that can effectively analyze temporal patterns in urban traffic, uncovering underlying trends and factors influencing traffic flow, which are essential for optimizing smart city infrastructure. To achieve this, we applied DBSCAN, K-Means, Affinity Propagation, Mean Shift, and Gaussian Mixture clustering techniques to the traffic dataset of Aarhus, Denmark's second-largest city. The performance of these clustering methods was evaluated using the Silhouette Score and Dunn Index, with DBSCAN emerging as the most effective algorithm in terms of cluster quality and computational efficiency. The study also compares the training times of the algorithms, revealing that DBSCAN, K-Means, and Gaussian Mixture offer faster training times, while Affinity Propagation and Mean Shift are more computationally intensive. The results demonstrate that DBSCAN not only provides superior clustering performance but also operates efficiently, making it an ideal choice for analyzing urban traffic patterns in large datasets. This research emphasizes the importance of selecting appropriate clustering techniques for effective traffic analysis and management within smart city frameworks, thereby contributing to more efficient urban planning and infrastructure development.

Author 1: Praveena Kumari M K

Author 2: Manjaiah D H

Author 3: Ashwini K M

Keywords: Clustering; smart city; traffic; analyze

Paper 12: Prediction of Outpatient No-Show Appointments Using Machine Learning Algorithms for Pediatric Patients in Saudi Arabia

Abstract: Patient no-shows are prevalent in pediatric outpatient visits, leading to underutilized medical resources, increased healthcare costs, reduced clinic efficiency, and decreased access to care. The use of machine learning techniques provides insights to mitigate this problem. This study aimed to develop a predictive model for patient no-shows at the Ministry of National Guard Health-Affairs, Saudi Arabia, and evaluate the results of various machine learning algorithms in predicting these events. Four machine learning algorithms - Gradient Boosting, AdaBoost, Random Forest, and Naive Bayes - were used to create predictive models for patient no-shows. Each model underwent extensive parameter tuning and reliability assessment to ensure robust performance, including sensitivity analysis and cross-validation. Gradient Boosting achieved the highest area under the receiver operating curve (AUC) of 0.902 and Classification Accuracy (CA) of 0.944, while the AdaBoost model achieved an AUC of 0.812 and CA of 0.927. The Naive Bayes and Random Forest models achieved AUCs of 0.677 and 0.889 and CAs of 0.915 and 0.937, respectively. The confusion matrix demonstrated high true-positive rates for no-shows for the Gradient Boosting and Random Forest models, while Naive Bayes had the lowest values. The Gradient Boosting and Random Forest models were most effective in predicting patient no-shows. These models could enhance outpatient clinic efficiency by predicting no-shows. Future research can further refine these models and investigate practical strategies for their implementation.

Author 1: Abdulwahhab Alshammari

Author 2: Fahad Alotaibi

Author 3: Sana Alnafrani

Keywords: No-show; pediatric; machine learning; algorithms; prediction; outpatients

Paper 13: Performance Optimization of Support Vector Machine with Adversarial Grasshopper Optimization for Heart Disease Diagnosis and Feature Selection

Abstract: The World Health Organization reports that cardiac disorders result in approximately 1.02 million deaths. Over the last years, heart disorders, also known as cardiovascular diseases, have significantly influenced the medical sector due to their immense global impact and high level of danger. Unfortunately, accurate prognosis of heart problems or CD, as well as continuous monitoring of the patient for 24 hours, is unattainable due to the extensive expertise and time required. The management and identification of cardiac disease pose significant challenges, particularly in impoverished or developing nations. Moreover, the absence of adequate medical attention or prompt disease management can result in the individual's demise. This study presents a novel optimization technique for diagnosing cardiac illness utilizing Support Vector Machine (SVM) and Grasshopper Optimization Algorithm (GOA). The primary objective of this approach is to identify the most impactful characteristics and enhance the efficiency of the SVM model. The GOA algorithm, which draws inspiration from the natural movements of grasshoppers, enhances the search for features in the data and effectively reduces the feature set while maintaining prediction accuracy. The initial stage involved pre-processing the ECG data, followed by its classification using several algorithms such as SVM and GOA. The findings demonstrated that the suggested approach has markedly enhanced the effectiveness and precision of heart disease diagnosis through meticulous feature selection and model optimization. This approach can serve as an efficient tool for early detection of heart disease by simplifying the process and enhancing its speed.

Author 1: Nan Tang

Author 2: Lele Wang

Author 3: Kangming Li

Author 4: Zhen Liu

Author 5: Yanan Dai

Author 6: Ji Hao

Author 7: Qingdui Zhang

Author 8: Huamei Sun

Author 9: Chunmei Qi

Keywords: Heart disease predictions; Support Vector Machine; Grasshopper Optimization Algorithm; feature selection

Paper 14: Sleep Disorder Diagnosis Through Complex-Morlet-Wavelet Representation Using Bi-GRU and Self-Attention

Abstract: Sleep disorders pose notable health risks, impacting memory, cognitive performance, and overall well-being. Traditional polysomnography (PSG) used for sleep disorder diagnosis are complex and inconvenient due to complex multi-class representation of signals. This study introduces an automated sleep-disorder-detection method using electrooculography (EOG) and electroencephalography (EEG) signals to address the gaps in automated, real-time, and noninvasive sleep-disorder diagnosis. Traditional methods rely on complex PSG analysis, whereas the proposed method simplifies the involved process, reducing reliance on cumbersome equipment and specialized settings. The preprocessed EEG and EOG signals are transformed into a two-dimensional time-frequency image using a complex-Morlet-wavelet (CMW) transform. This transform assists in capturing both the frequency and time characteristics of the signals. Afterwards, the features are extracted using a bidirectional gated recurrent unit (Bi-GRU) with a self-attention layer and an ensemble-bagged tree classifier (EBTC) to correctly classify sleep disorders and very efficiently identify them. The overall system combines EOG and EEG signal features to accurately classify people with insomnia, narcolepsy, nocturnal frontal lobe epilepsy (NFLE), periodic leg movement (PLM), rapid-eye-movement (RBD), sleep behavior disorder (SDB), and healthy, with success rates of 99.7%, 97.6%, 95.4%, 94.5%, 96.5%, 98.3%, and 94.1%, respectively. Using the 10-fold cross-validation technique, the proposed method yields 96.59% accuracy and AUC of 0.966 with regard to classification of sleep disorders into multistage classes. The proposed system assists medical experts for automated sleep-disorder diagnosis.

Author 1: Mubarak Albathan

Keywords: Deep learning; complex morlet wavelet; bidirectional gated recurrent unit; sleep stage detection; multistage sleep disorder; ensemble-bagged tree classifier

Paper 15: Digital Landscape Architecture Design Combining 3D Image Reconstruction Technology

Abstract: To achieve better digital landscape design and visual presentation effects, this study proposes a digital landscape design method based on improved 3D image reconstruction technology. Firstly, a precise point cloud registration algorithm combining normal distribution transformation and Trimmed iterative nearest point algorithm is proposed. A color texture method for 3D models is designed in terms of 3D reconstruction, and a visual scene, 3D reconstruction method based on RGBD data is constructed. Secondly, knowledge networks are introduced to assist in the intelligent generation and planning of plant communities in urban landscape scenes. The knowledge network established through the plant database integrates the principles of landscape design and optimizes the layout of landscape plants in urban parks. The running speed and accuracy of research algorithms were superior to traditional methods, especially in terms of registration performance. Compared to the other two algorithms, the registration time of the research algorithm was reduced by 2%, and the errors were reduced by 71.4% and 87.5%, respectively. The panoramic quality of research methods fluctuated within a small range of 0.8 or above, while traditional methods exhibited instability and lower quality. The landscape design generated by research methods was more aesthetically pleasing and harmonious with the actual landscape in terms of plant selection and layout. The proposed method follows the principles of eco-friendly design and demonstrates significant potential for application in the field of urban landscape design.

Author 1: Chen Chen

Keywords: 3D image reconstruction; PSO; gardens; RGBD; digital landscape

Paper 16: Software Systems Documentation: A Systematic Review

Abstract: In the domain of software engineering, software documentation encompasses the methodical creation and management of artifacts describing software systems. Traditionally linked to software maintenance, its significance extends throughout the entire software development lifecycle. While often regarded as a quintessential indicator of software quality, the perception of documentation as a time-consuming and arduous task frequently leads to its neglect or obsolescence. This research presents a systematic review of the past decade's literature on software documentation to identify trends and challenges. Employing a rigorous systematic methodology, the study yielded 29 primary studies and a collection of related works. Analysis of these studies revealed two primary themes: issues and best practices, and models and tools. Findings indicate a notable research gap in the area of software documentation. Furthermore, the study underscores several critical challenges, including a dearth of automated tools, immature documentation models, and an insufficient emphasis on forward-looking documentation.

Author 1: Abdullah A H Alzahrani

Keywords: Software engineering; software systems documentation; software maintenance; software quality; software development

Paper 17: Optimizing Data Security in Computer-Assisted Test Applications Through the Advanced Encryption Standard 256-Bit Cipher Block Chaining

Abstract: In the digital education era, the importance of Computer-Assisted Test programs is underscored by their efficiency in conducting assessments. However, the increasing incidence of data breaches and cyberthreats has made the implementation of robust data protection measures imperative. This study explores the adoption of the Advanced Encryption Standard 256-bit Cipher Block Chaining in CAT applications to enhance data security. Known for its strong encryption capabilities, AES-256-CBC is an excellent choice for securing sensitive test data. The research focuses on the application of AES-256-CBC within CAT systems during the independent admission process at Politeknik Negeri Bengkalis, a critical phase where the integrity of exam materials and student data is paramount. We evaluate the effectiveness of AES-256-CBC in encrypting user data and exam materials across different CAT systems, thus preserving data integrity and confidentiality. The implementation of AES-256-CBC helps prevent unauthorized access and manipulation of test results, ensuring a secure online testing environment. This research not only demonstrates the technical implementation of AES-256-CBC but also assesses its impact on enhancing the security posture of CAT applications at Politeknik Negeri Bengkalis. The findings contribute to the broader discussion on data security in educational technology, positioning AES-256-CBC as a potent solution for maintaining academic integrity in digital testing environments.

Author 1: M. Afridon

Author 2: Agus Tedyyana

Author 3: Fajar Ratnawati

Author 4: Afis Julianto

Author 5: M. Nur Faizi

Keywords: AES256-CBC; data security; computer-assisted test; academic integrity; encryption standards; digital assessment security

Paper 18: Diabetes Prediction Using Machine Learning with Feature Engineering and Hyperparameter Tuning

Abstract: Diabetes, a chronic illness, has seen an increase in prevalence over the years, posing several health challenges. This study aims to predict diabetes onset using the Pima Indians Diabetes dataset. We implemented several machine learning algorithms, namely Random Forest, Gradient Boosting, XGBoost, LightGBM, and CatBoost. To enhance model performance, we applied a variety of feature engineering techniques, including SelectKBest, Recursive Feature Elimination (RFE), Recursive Feature Elimination with Cross-Validation (RFECV), Forward Feature Selection, and Backward Feature Elimination. RFECV proved to be the most effective method, leading to the selection of the best feature set. In addition, hyperparameter tuning techniques are used to determine the optimal parameters for the models created. Upon training these models with the optimized parameters, XGBoost outperformed the others with an accuracy of 94%, while Random Forest and CatBoost both achieved 92.5%. These results highlight XGBoost's superior predictive power and the significance of thorough feature engineering and model tuning in diabetes prediction.

Author 1: Hakim El Massari

Author 2: Noreddine Gherabi

Author 3: Fatima Qanouni

Author 4: Sajida Mhammedi

Keywords: Machine learning; feature engineering; hyperparameter tuning; diabetes prediction; healthcare

Paper 19: Computational Modeling of the Thermally Stressed State of a Partially Insulated Variable Cross-Section Rod

Abstract: The formulation of the proposed methods and algorithms facilitates a comprehensive examination of intricate non-stationary thermo-mechanical processes in rods with varying cross-sectional geometries. Furthermore, it advances the theoretical framework for analyzing the thermo-mechanical properties of rod structures utilized in the machinery industry of the Republic of Kazakhstan. The creation of these intellectual products aids in the progression of this sector and fortifies the nation's sovereignty. This article delineates methods and algorithms for investigating non-stationary thermo-mechanical processes in rods with diverse cross-sectional shapes that influence global manufacturing technologies. The scientific and practical importance of this work lies in the application potential of the developed approach for examining non-stationary thermo-mechanical characteristics of rod-like elements in various installations. The findings also enhance the scientific research direction in mechanical engineering. In conclusion, the article outlines future technological advancements, summarizes the research on non-stationary thermo-mechanical processes in rods with different cross-sectional geometries, and highlights significant economic benefits by facilitating the selection of reliable rods for specified operating conditions. This ensures the continuous and dependable operation of machinery used in mechanical engineering.

Author 1: Zhuldyz Tashenova

Author 2: Elmira Nurlybaeva

Author 3: Zhanat Abdugulova

Author 4: Shirin Amanzholova

Author 5: Nazira Zharaskhan

Author 6: Aigerim Sambetova

Author 7: Anarbay Kudaykulov

Keywords: Heat flow; heat transfer; thermal expansion coefficient; thermal conductivity; modulus of elasticity

Paper 20: Performance Analysis of a Hyperledger-Based Medical Record Data Management Using Amazon Web Services

Abstract: Recently, there's been growing excitement around the innovative capabilities of blockchain technology, especially for enhancing security, privacy, and transparency. Its application in various sectors, like finance and logistics, is intriguing, but its potential in healthcare stands out. Specifically, in the realm of medical data management, blockchain can transform how we protect patient data. Our study unveils a cutting-edge approach to handle digital health records by harnessing the power of Amazon Web Services (AWS). This pioneering, serverless model is not only cost-effective, with charges only for used resources, but also offers heightened security and for blockchain network access. We build a private, permissioned blockchain network with Hyperledger Fabric to control access while ensuring transparency. The paper demonstrates the prowess of this new system is validated through rigorous tests on speed, network prowess, and multi-user handling, complete with a detailed cost analysis for implementation. The paper further demonstrates the use of the Gatling open-source library to design various experiments for performance measurement.

Author 1: Mohammed K Elghoul

Author 2: Sayed F. Bahgat

Author 3: Ashraf S. Hussein

Author 4: Safwat H. Hamad

Keywords: Hyperledger; blockchain; healthcare; data management

Paper 21: Enhancing Business Intelligence with Hybrid Transformers and Automated Annotation for Arabic Sentiment Analysis

Abstract: Business is a key focus for many individuals, companies, countries and organisations. One effective way to enhance business performance is by analysing customer opinions through sentiment analysis. This technique offers valuable insights, known as business intelligence, which directly benefits business owners by informing their decisions and strategies. Substantial attention has been given to business intelligence through proposed machine learning approaches, deep learning models and approaches utilizing natural language processing methods. However, building a robust model to detect and identify users’ opinion and automated text annotation, particularly for the Arabic language, still faces many challenges. Thus, this study aims to propose a hybrid transfer learning model that uses transformers to identify positive and negative user comments that are related to business. This model consists of three pretrained models, namely, AraBERT, ArabicBERT, and XLM-RoBERTa. In addition, this study proposes a hybrid automatic Arabic annotation method based on CAMelBERT, TextBlob and Farasa to automatically classify user comments. A novel dataset, which is collected from user-generated comments (i.e. reviews on mobile apps), is introduced. This dataset is annotated twice using the proposed method and human-based annotation. Then, several experiments are conducted to evaluate the performance of the proposed model and the proposed annotation method. Experiment results show that the proposed hybrid model outperforms the baseline models, and the proposed annotation method achieves high accuracy, which is close to human-based annotation.

Author 1: Wael M.S. Yafooz

Keywords: Business intelligence; machine learning; sentiment analysis; transformers; BERT; Arabic annotation

Paper 22: A Method by Utilizing Deep Learning to Identify Malware Within Numerous Industrial Sensors on IoTs

Abstract: The industrial sensors of IoT is an emerging model, which combines Internet and the industrial physical smart objects. These objects belong to the broad domains like the smart homes, the smart cities, the processes of the industrial and the military, the agriculture and the business. Due to the substantial advancement in Industrial Internet of Things (IIoT) technologies, numerous IIoT applications have been developed over the past ten years. Recently, there have been multiple reports of malware-based cyber-attacks targeting IIoT systems. Consequently, this research focuses on creating an effective Artificial Intelligence (AI)-powered system for detecting zero-day malware in IIoT environments. In the current article, a combined framework for the detection of the malware basis on the deep learning (DL) is proposed, that uses the dual-density discrete wavelet transform for the extraction of the feature and a combination from the convolutional neural network (CNN) and the long-term short-term memory (LSTM). The method is utilized for malware detection and classification. It has been assessed using the Malimg dataset and the Microsoft BIG 2015 dataset. The results demonstrate that our proposed model can classify malware with remarkable accuracy, surpassing similar methods. When tested on the Microsoft BIG 2015 and Malimg datasets, the accuracy achieved is 95.36% and 98.12%, respectively.

Author 1: Ronghua MA

Keywords: Malware; malware detection; industrial sensors; Internet of Things (IoTs); Deep Learning (DL)

Paper 23: Quantifying the Effects of Homogeneous Interference on Coverage Quality in Wireless Sensor Networks

Abstract: This study develops a coverage perception interference model for Wireless Sensor Networks, focusing on the challenges of homogeneous interference within Regions of Interest. Traditional perception models often overlook areas that, while covered, do not meet the required coverage standards for accurate classification. This model addresses both uncovered areas and those inadequately covered, which are susceptible to classification errors. A propositional space for the coverage model is defined to assess the impact of homogeneous interference on sensor nodes, with the aim of quantifying its effects on network coverage quality and stability in complex environments. The study emphasizes the generation of Basic Probability Assignments using Dempster-Shafer theory, a robust framework for managing uncertain information in sensory data. Probability Density Functions derived from historical and real-time data are utilized to facilitate precise BPA calculations by integrating over specific attribute ranges, thereby enhancing the accuracy and reliability of target detection. Algorithms are also developed to calculate the interference effect BPA, which are integrated with perception coverage models to improve the assessment and optimization of coverage quality. The research enhances the methodological understanding of managing interference in WSNs and offers practical strategies for improving sensor network operations in environments affected by significant interference, boosting the reliability and effectiveness of critical surveillance and monitoring applications.

Author 1: Qingmiao Liu

Author 2: Qiang Liu

Author 3: Minhuan Wang

Keywords: Wireless sensor networks; homogeneous interference; basic probability assignment; coverage quality; Dempster-Shafer theory

Paper 24: Lightweight and Efficient High-Resolution Network for Human Pose Estimation

Abstract: To address the challenges of high parameter quantities and elevated computational demands in high-resolution network, which limit their application on devices with constrained computational resources, we propose a lightweight and efficient high-resolution network, LE-HRNet. Firstly, we designs a lightweight module, LEblock, to extract feature information. LEblock leverages the Ghost module to substantially decrease the number of model parameters. Based on this, to effectively recognize human keypoints, we designed a Multi-Scale Coordinate Attention Mechanism (MCAM). MCAM enhances the model's perception of details and contextual information by integrating multi-scale features and coordinate information, improving the detection capability for human keypoints. Additionally, we designs a Cross-Resolution Multi-Scale Feature Fusion Module (CMFFM). By optimizing the upsampling and downsampling processes, CMFFM further reduces the number of model parameters while enhancing the extraction of cross-branch channel features and spatial features to ensure the model's performance. The proposed model's experimental results demonstrate accuracies of 69.3% on the COCO dataset and 88.7% on the MPII dataset, with a parameter count of only 5.4M, substantially decreasing the number of model parameters while preserving its performance.

Author 1: Jiarui Liu

Author 2: Xiugang Gong

Author 3: Qun Guo

Keywords: Human pose estimation; model lightweighting; Ghost module; attention mechanism; multi-scale feature fusion

Paper 25: Enhanced Resume Screening for Smart Hiring Using Sentence-Bidirectional Encoder Representations from Transformers (S-BERT)

Abstract: In a world inundated with resumes, the hiring process is often challenging, particularly for large organizations. HR professionals face the daunting task of manually sifting through numerous applications. This paper presents ‘Enhanced Resume Screening for Smart Hiring using Sentence-Bidirectional Encoder Representations from Transformers (S-BERT)’ to revolutionize this process. For HR professionals dealing with overwhelming numbers of resumes, the manual screening process is time consuming and error-prone. To address this, here the proposed solution is developed for an automated solution leveraging NLP techniques and a cosine distance matrix. Our approach involves pre-processing, embed- ding generation using S-BERT, cosine similarity calculation, and ranking based on scores. In our evaluation on a dataset of 223 resumes, our automated screening mechanism demonstrated remarkable efficiency with a screening speed of 0.233 seconds per resume. The system’s accuracy was 90%, showcasing its ability to effectively identify relevant resumes. This work presents a powerful tool for HR professionals, significantly reducing the manual workload and enhancing the accuracy of identifying suitable candidates. The societal impact lies in streamlining hiring processes, making them more efficient and accessible, ultimately contributing to a more productive and equitable job market.

Author 1: Asmita Deshmukh

Author 2: Anjali Raut

Keywords: S-BERT; resume; automated screening; job; CV

Paper 26: Machine Learning Techniques for Protecting Intelligent Vehicles in Intelligent Transport Systems

Abstract: Intelligent transport system (ITS) is the development direction of future transport systems, in which intelligent vehicles are the key components. In order to protect the safety of intelligent vehicles, machine learning techniques are widely used in ITS. For intelligent protection in ITS, the study introduces an improved driving behaviour modelling method based on Bagging Gaussian Process Regression. Meanwhile, to further promote the accuracy of driving behaviour modelling and prediction, Convolutional Neural Network-Long and Short-term Memory Network-Gaussian Process Regression are used for effective feature extraction. The results show that in the straight overtaking scenario, the mean absolute error, root mean square error and maximum absolute error of the improved Bagging Gaussian process regression method are 0.5241, 0.9547 and 10.7705, respectively. In the corner obstacle avoidance scenario, the improved Bagging Gaussian process regression method is only 0.6527, 0.9436 and 14.7531. Besides, the mean absolute error of the Convolutional Neural Network-Long and Short-term Memory Network-Gaussian process regression algorithm is only 0.0387 in the case of the input temporal image frame number of 5. This denoted that the method put forward in the study can provide a more accurate and robust modeling and prediction of driving behaviours in complex traffic environments, and it has a high application potential in the field of safety and protection of intelligent vehicles.

Author 1: Yuan Chen

Keywords: Intelligent vehicle protection; machine learning techniques; Gaussian Process Regression; convolutional neural networks; long and short-term memory networks

Paper 27: Automation of Book Categorisation Based on Network Centric Quality Management System

Abstract: In order to improve the efficiency of automatic book classification, the study uses a crawler to crawl book data from regular websites and perform data cleaning and fusion to build a structured knowledge graph. Meanwhile, the processed data is applied to a pre-trained model to improve it, and migration learning is used to improve the results. Fusion of Multiple Attention Mechanisms, Recurrent Neural Network, and Convolutional Neural Network modules into the classification model and feature fusion is used to enhance feature extraction. In addition, the study designed a pre-trained model architecture to help automatically categorise and manage book resources. The results of this study show a significant improvement in the classification of Chinese books on the Chinese Book L2 Subject Classification, iFlytek, and THUCNews datasets with significant performance improvement. The fusion of long and short-term memory and convolutional network Transformer-based bi-directional encoding models improved the accuracy by 0.19%, 1.54% and 0.42% on the two datasets, respectively, while the weighted average F1 scores improved accordingly. Through wireless technology, the automatic classification efficiency of books is realized and the management ability of the library is improved.

Author 1: Tingting Liu

Author 2: Qiyuan Liu

Author 3: Linya Fu

Keywords: Crawler; books; automated; classification; Recurrent Neural Network; Multiple Attention Mechanism; knowledge graphs

Paper 28: Optimization of Distribution Routes in Agricultural Product Supply Chain Decision Management Based on Improved ALNS Algorithm

Abstract: The transportation of fresh agricultural products is not conducted along a sufficiently precise route, resulting in an extended transportation time for vehicles and a consequent deterioration in product freshness. Therefore, the study proposes an agricultural product transportation path optimization model based on an optimized adaptive large neighborhood search algorithm. The Solomon standard test case is used for the experiment, and the algorithm before and after optimization is compared. From the results, the optimized method was effective for the distribution model C201, R201, and CR201 sets after conducting case analysis. The total cost of the R201 transportation set was the lowest, while C101 had the highest total cost. The lowest vehicle cost consumption was R201 at 600, and the highest was C101 at 2220. The C101 algorithm took 145 s to calculate, and R201 took 199 s. All values of CR201 were average, with high fault tolerance. The proposed method was used to address the optimal operator solution. The C201 example took 244 s to calculate 2350 objective function values. The R201 example took 239 s to obtain 657 objective function values. The CR201 example took 233 s to obtain 764 objective function values. This indicates that the designed method has a significant effect on optimizing the distribution path of agricultural products. Compared with the unimproved algorithm, it has more accurate search ability and lower transportation costs. This algorithm provides path optimization ideas for the agricultural product transportation industry.

Author 1: Liling Liu

Author 2: Yang Chen

Author 3: Ao Li

Keywords: ALNS; agricultural products; path optimization; cold chain transportation; supply chain

Paper 29: Harnessing Technology to Achieve the Highest Quality in the Academic Program of University Studies

Abstract: This research aims to utilize information technology to improve education quality, particularly in higher education. A key contribution of this research is the application of generative artificial intelligence, specifically ChatGPT, to validate test questions that meet both international (ABET) and local (NCAAA) academic accreditation standards. The study was conducted within the Information Systems Department's bachelor's program at King Abdulaziz University in the Kingdom of Saudi Arabia, focusing on a website development course. The custom ChatGPT application, named Question Checker, was developed to validate questions generated by instructors. These validation criteria were aligned with the accreditation requirements for technology and computer science programs, ensuring compliance with both ABET and NCAAA standards. The application was tested by validating nine questions related to Student Outcomes, demonstrating its effectiveness in supporting the educational objectives of the program.

Author 1: Rania Aboalela

Keywords: ChatGPT; academic accreditations; technology programs; computer science; Kingdom of Saudi Arabia; website development; NCAAA; ABET

Paper 30: Enhancing Digital Financial Security with LSTM and Blockchain Technology

Abstract: The growing dependence on digital financial and banking transactions has brought about a significant focus on implementing strong security protocols. Blockchain technology has proved itself throughout the years to be a reliable solution upon which transactions can safely take place. This study explores the use of blockchain technology, specifically Ethereum Classic (ETC), to enhance the security of digital financial and banking transactions. The aim is to develop a system using an LSTM model to predict and detect anomalies in transaction data. The proposed LSTM model was trained before being tested and the results prove that the proposed model can effectively enhance the security, especially when compared to other studies in the same domain. The proposed model achieved a prediction accuracy of 99.5%, demonstrating its effectiveness in enhancing security by preventing overfitting and identifying potential threats in network activities. The results suggest significant improvements in digital transaction security, enhancing both the traceability and transparency of blockchain transactions while reducing fraud rates. Future work will extend this model's applicability to larger-scale decentralized finance systems.

Author 1: Thanyah Aldaham

Author 2: Hedi HAMDI

Keywords: Digital financing; block chain; ETC; security; anomaly detection; machine learning; LSTM

Paper 31: Sketch and Size Orient Malicious Activity Monitoring for Efficient Video Surveillance Using CNN

Abstract: Towards malicious activity monitoring in organizations, there exist several techniques and suffer with poor accuracy. To handle this issue, an efficient Sketch and Size orient malicious activity monitoring (SSMAM) is presented in this article. The model captures the video frames and performs segmentation to extract the features of frames as shapes and size. The video frames are enhanced for its quality by applying High Level Intensity Analysis algorithm. The quality improved image has been segmented with Color Quantization Segmentation. Using the segmented image, the feature are extracted and applied with scaling and rotation for different number of size and angle. Such features extracted have been trained with convolution neural network. The CNN model is designed to perform convolution on two levels and pooling as well. At the test phase, the method extract the same set of features and performs convolution to obtain same set of feature lengths and the neurons are designed computes the value of Sketch Support Measure (SSM) towards various class of activity. According the value of SSM, the method classifies the user activity towards efficient video surveillance. The proposed approach improves the performance in activity monitoring and video surveillance.

Author 1: K. Lokesh

Author 2: M. Baskar

Keywords: Video surveillance; deep learning; activity monitoring; malicious activity; SSMAM; SSM

Paper 32: Enhancing Arabic Phishing Email Detection: A Hybrid Machine Learning Based on Genetic Algorithm Feature Selection

Abstract: Recently, owing to widespread Internet use and technological breakthroughs, cyber-attacks have increased. One of the most common types of attacks is phishing, which is executed through email and is a leading cause of the recent surge in cyber-attacks. These attacks maliciously demand sensitive or private information from individuals and companies. Various methods have been employed to address this issue by classifying emails, such as feature-based classification and manual verification. However, these methods face significant challenges regarding computational efficiency and classification precision. This work presents a novel hybrid approach that combines machine learning and deep learning techniques to improve the identification of phishing emails containing Arabic content. A genetic algorithm is employed to optimize feature selection, thereby enhancing the performance of the model by effectively identifying the most relevant features. The novel dataset comprises 1,173 records categorized into two classes: phishing and legitimate. A number of empirical investigations were carried out to assess and contrast the performance outcomes of the proposed model. The findings reveal that the proposed hybrid model outperforms other machine learning classifiers and standalone deep learning models.

Author 1: Amjad A. Alsuwaylimi

Keywords: Machine learning; phishing email; BiLSTM; Arabic content-based

Paper 33: A Feature Interaction Based Neural Network Approach: Predicting Job Turnover in Early Career Graduates in South Korea

Abstract: Predicting job turnover among early career university graduates is crucial for both employees and employers. This study introduced a Feature Interaction based Neural Network model designed to predict job turnover among university graduates in their 20s and 30s in South Korea within the first five years of employment. The FINN model leveraged the Graduates Occupational Mobility Survey dataset, which included detailed information on approximately 26,544 graduates. This rich dataset encompassed a wide range of variables, including personal attributes, employment characteristics, job satisfaction, and job preparation activities. The model combined an embedding layer to convert sparse features into dense vectors with a neural network component to capture high-order feature interactions. We compared the FINN model's performance against eight baseline models: Logistic Regression, Factorization Machines, Field-aware Factorization Machines, Support Vector Machine, Random Forest, Product-based Neural Networks, Wide & Deep, and DeepFM. Evaluation metrics used were Area Under the ROC Curve (AUC) and Log Loss. The results demonstrated that the FINN model outperformed all baseline models, achieving an AUC of 0.830 and a Log Loss of 0.370. The FINN model represents a significant advancement in predictive modeling for job turnover, providing valuable insights that can inform both individual career planning and organizational human resource practices. This research underscores the potential of advanced neural network architectures in employment data analysis and predictive modeling.

Author 1: Haewon Byeon

Keywords: Job turnover prediction; feature interaction based neural network; employment data analysis; predictive modeling; university graduates

Paper 34: A Systematic Review of Virtual Commerce Solutions for the Metaverse

Abstract: The metaverse, a rapidly evolving field, promises to transform online shopping through immersive technologies. This systematic review aims to explore and analyze the key design features of Virtual Commerce (v-commerce) solutions within this digital environment. By examining 24 studies that have developed immersive v-commerce applications, this review seeks to compile a taxonomy of essential design attributes necessary for creating effective and engaging v-commerce experiences. The review classifies these attributes into three primary dimensions: Product, Intelligent Services, and Functionality. The findings indicate that within the Augmented Reality (AR) category, product visualization and natural interaction were the most studied attributes. In the Virtual Reality (VR) category, intuitive affordances emerged as the most frequently investigated features. Meanwhile, Mixed Reality (MR) studies commonly focused on information quality, intuitive affordances, and shopping assistants. The insights from this review provide valuable guidance for researchers, developers, and practitioners aiming to enhance consumer engagement and satisfaction in the metaverse through well-designed v-commerce applications. By synthesizing the results of various studies, this review offers a comprehensive overview of the current state of v-commerce research, identifies existing gaps, and proposes potential directions for future development in the field.

Author 1: Ghazala Bilquise

Author 2: Khaled Shaalan

Author 3: Manar Alkhatib

Keywords: Metaverse; v-commerce; immersive technologies; design attributes

Paper 35: DIAUTIS III: A Fuzzy and Affective Platform for Obtaining Autism Mental Models and Learning Aids

Abstract: Autism spectrum disorders (ASD) are conditions characterized by social interaction and communication difficulties, atypical patterns of activities, and unusual reactions to sensations. Char-acteristics of autism may be detected in early childhood, but diagnosis is often delayed. The diagnosis of autistic children typically aligns with medical and psychological recommenda-tions, but it does not evaluate all the problems, intensity, or changes in symptoms over time. It also does not identify the affective states associated with these deficiencies, making aid less effective. The mental model of autistic children contains their deficits, tasks, and intensities, beyond diagnostics. That why we enhance DIAUTIS platform for achieve our objectives related to helping children with ASD. DIAUTIS I is a platform that aim to diagnosing autism and identifying its severity using cognitive, fuzzy, and affective computing. It presents tests, eval-uates results, and presents a final model. Then, we implemented DIAUTIS II by adding KASP methodology, a new methodology of designing serious games, based on knowledge, affect, sensory, and pedagogy, this tool allows to DIAUTIS II agents to design-ing over 80 games, considering a child's background. In this paper, we will present a new tool of formalization of the autism mental model based on fuzzy and affective computing. DIAUTIS III is the extension of DIAUTIS II platform aim to represent a cognitive fuzzy mental model with using the metrics of category theory. So far, no mental model has been developed for autism. Our mental model, can be obtained anytime as a fuzzy cognitive map or fuzzy graph and the use of affective computing. In addi-tion, the mathematical theory of categories represent this men-tal model of an autistic child from a fuzzy graph, and it allows for operations like CONS and TRA to evaluate the difference between two mental models. Fuzzy cognitive mental model can be used to develop new techniques for improving the learning and integration of autistic children into social life, which is the focus of our immediate future.

Author 1: Mohamed El Alami

Author 2: Sara El khabbazi

Author 3: Fernando de Arriaga

Keywords: Fuzzy computing; affective computing; mental models; learning aids; autism; category theory

Paper 36: Stock Price Forecasting with Optimized Long Short-Term Memory Network with Manta Ray Foraging Optimization

Abstract: The stock market is a financial marketplace where investors may participate through the acquisition and sale of stocks in publicly traded companies. Predicting stock prices in the securities sector may be challenging due to the intricate nature of the subject, which necessitates a comprehensive grasp of several interconnected factors. Numerous factors, including politics, society, as well as the economy, have an impact on the stock market. The primary objective of financial market investing is to exploit larger profits. Financial markets provide many opportunities for market analysts, investors, and researchers in several industries due to significant technology advancements. Conventional approaches encounter difficulties in capturing the complex, non-linear connections that exist in market data, which requires the implementation of sophisticated artificial intelligence models. This paper presents a new approach to tackling certain issues by suggesting a unique model. It combines the long short-term memory method and Empirical Mode Decomposition with the Manta Ray Foraging Optimization. When tested in the current study's dynamic stock market, the EMD-MRFO-LSTM model outperformed other models regarding performance and efficiency. The Nasdaq index data from January 2, 2015, to June 29, 2023, were used in this study. The findings demonstrate how the suggested model is capable of making precise stock price predictions. The suggested model offers a workable approach to studying and predicting stock price time series by obtaining values of 0.9973, 91.99, 71.54, and 0.57, for coefficient of determination (R^2), root means square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), respectively. Compared to other methods currently in use, the proposed model has a higher accuracy in forecasting and is more physically relevant to the dynamic stock market, according to the outcomes of the experiment.

Author 1: Zhongpo Gao

Author 2: Junwen Jing

Keywords: Stock price; hybrid forecasting method; Manta Ray Foraging Optimization; empirical mode decomposition; Nasdaq index

Paper 37: Modified TOPSIS Method for Neutrosophic Cubic Number Multi-Attribute Decision-Making and Applications to Music Composition Effectiveness Evaluation of Film and Television

Abstract: Contemporary music composition for film and television has exhibited a trend towards diversification, which is reflected in various aspects such as the diversity of musical styles, the integration of music and visuals, as well as the technical means of music creation. With the continuous advancement of music production technology and film/television production technology, the creation of music for film and television has increasingly emphasized the organic integration of music and visuals, as well as the role of music in emotional expression and atmosphere creation. Meanwhile, the fusion and innovation of different musical styles have also brought more possibilities and space to the creation of music for film and television. This trend of diversification not only enriches the artistic expressiveness of film and television works, but also provides audiences with a more diverse audiovisual experience. The music composition effectiveness evaluation of film and television is multi-attribute decision-making (MADM) problem. In this paper, the TOPSIS method is extended to the framework of neutrosophic cubic sets (NCSs) to address MADM problems. The CRITIC method is employed to obtain the weights of the attributes, ensuring a systematic and objective approach to determining their relative importance. Furthermore, the neutrosophic cubic number TOPSIS (NCN-TOPSIS) approach is established for MADM scenarios. To demonstrate the applicability of the proposed NCN-TOPSIS model, a numerical example focused on the music composition effectiveness evaluation in film and television is presented. Additionally, comparative analyses are conducted to showcase the advantages of the NCN-TOPSIS approach over other decision-making methods. By extending the TOPSIS technique to the NCSs environment and integrating the CRITIC method, this research contributes to the field of MADM by providing a robust and efficient decision-making tool for complex, multi-criteria problems, as exemplified by the music composition evaluation in the film and television industry.

Author 1: Liang Yang

Author 2: Jun Zhao

Keywords: Multi-Attribute Decision-Making (MADM); NCSs; TOPSIS approach; music composition effectiveness evaluation

Paper 38: Heuristic Intelligent Algorithm-Based Approach for In-Depth Development and Application Analysis of Micro- and Nanoembedded Systems

Abstract: Developing application analysis and testing methods is an important part of the in-depth development of micro-nano embedded systems under complex integrated architectures. Therefore, the research on application analysis and testing models is of great significance for the in-depth development and efficient design of embedded systems. In order to fully explore the effective information of test analysis data in the in-depth development of micro-nano embedded systems under complex integrated architectures and improve the analysis and prediction accuracy of test analysis models, a development test analysis model based on the photon search algorithm and LightGBM is proposed. First, the development process of micro-nano embedded systems under complex integrated architectures is analysed, and a system analysis architecture is designed to propose test analysis factors. Second, a development test analysis model is established by combining the photon search algorithm and LightGBM. Subsequently, the feasibility and efficiency of the model are proposed by analysing the data of the development process. The analysis of data examples shows that the LightGBM test analysis model has high analysis and prediction accuracy and generalisation performance.

Author 1: Buzhong Liu

Keywords: Photon search algorithm; deep development of micro- and nanoembedded systems; application test and analysis methodology; LightGBM

Paper 39: Optimization of Knitting Path of Flat Knitting Machine Based on Reinforcement Learning

Abstract: In the textile industry, the flat knitting machine plays a crucial role as a production tool, and the quality of its weaving path is closely related to the overall product quality and production efficiency. Seeking to improve and optimize the knitting path to improve product effectiveness and productivity has become an urgent concern for the textile industry. This article elegantly streamlines and enhances the intricate weaving process of fabrics, harnessing the formidable power of reinforcement learning to achieve unparalleled optimization of weaving paths on a flat knitting machine. By ingeniously integrating reinforcement learning technology into the fabric production realm, we aspire to elevate both the quality and production efficiency of textiles to new heights. The core of our approach lies in meticulously defining a state space, action space, and a tailored reward function, each meticulously crafted to mirror the intricacies of the knitting process. This model serves as the cornerstone upon which we construct an innovative knitting pathway optimization algorithm, deeply rooted in the principles of reinforcement learning. Our algorithm embodies a relentless pursuit of excellence, learning from its interactions with the dynamic environment, embracing a methodical trial-and-error approach, and continuously refining its decision-making strategy. Its ultimate goal: to maximize the long-term cumulative reward, ensuring that every stitch contributes to the overall optimization of the weaving process. In essence, we have forged a groundbreaking collaboration between the traditional art of fabric weaving and the cutting-edge science of reinforcement learning, ushering in a new era of intelligent and efficient textile production. Through this process of iterative optimization, the agent can gradually learn the optimal knitting path. To verify the effectiveness of the algorithm, we performed extensive experimental validation. The experimental results show that reinforcement learning can significantly improve knitting efficiency, improve the appearance and feel of fabrics. Compared with traditional methods, the method proposed in this article has a higher level of automation and better adaptability, achieving more efficient and intelligent knitting production, with a 10% increase in production efficiency.

Author 1: Tianqi Yang

Keywords: Flat knitting machine; reinforcement learning; weaving path optimization; textile industry

Paper 40: Design and Research of Cross-Border E-Commerce Short Video Recommendation System Based on Multi-Modal Fusion Transformer Model

Abstract: This study designed a cross-border e-commerce short video recommendation system based on Transformer's multimodal analysis model. When mining associations, the model not only focuses on the relationships between modalities, but also improves semantic context by addressing contextual correlations within and between modalities. At the same time, the model uses a cross modal multi head attention mechanism for multi-level association mining, and constructs an association network interwoven with latitude and longitude. In the process of exploring the essential correlation between patterns and subjective emotional fluctuations, the potential context between patterns has been realized. Fully explore correlations and then more accurately identify the truth contained in the original data. In addition, this study proposes a self supervised single modal label generation method. When multimodal labels are known, it does not require complex deep networks and only relies on the mapping relationship between multimodal representations and labels to generate a single modal label. Modal labeling can achieve phased automatic labeling of single modal labels, and quantify the mapping relationship between modal representations and labels from the representation space to generate weak single modal labels. The study also achieved multimodal collaborative learning in the context of limited differential information acquisition due to incomplete labeling, fully utilizing multimodal information. The experimental results on classic datasets in the field of multimodal analysis show that it outperforms the baseline model in terms of accuracy and F1 score, reaching 98.76% and 97.89%, respectively.

Author 1: Yiran Hu

Keywords: Multimodal fusion; transformer model; cross-border e-commerce; short video recommendation system

Paper 41: A Hidden Markov Model-Based Performance Recognition System for Marching Wind Bands

Abstract: This paper explores the automatic recognition of marching band performances using advanced music information retrieval techniques. Music, a crucial medium for emotional expression and cultural exchange, greatly benefits from the harmonic backing provided by marching wind orchestras. Identifying these performances manually is both time-consuming and labor-intensive, particularly for non-professionals. This study addresses this challenge by leveraging Hidden Markov Models (HMM) and improved Pitch Class Profile (PCP) features to automate the recognition process. The research also explores the system's performance on real-world audio recordings with background noise and microphone variations. By dividing the audio signal into frames and transforming it to the frequency domain, the PCP feature vectors are extracted and used within the HMM framework. Experimental results demonstrate that the proposed method significantly enhances recognition accuracy compared to traditional PCP features and template matching models. The study identifies challenges in distinguishing similar tonal values, such as F-major and D-minor, which affect recognition rates. Additionally, the research highlights the importance of addressing background noise and microphone variations in real-world applications. Ethical considerations regarding privacy and intellectual property rights are also discussed. This research establishes a comprehensive system for automatic marching band performance recognition, contributing to advancements in music information retrieval and analysis.

Author 1: Wei Jiang

Keywords: Music information retrieval; Hidden Markov Model; feature extraction; automatic music recognition; marching band performance; PCP features

Paper 42: Fitness Equipment Design Based on Web User Text Mining

Abstract: To propose home fitness equipment that meets modern users' needs, this study employs web user text mining, combined with the Fuzzy Analytic Hierarchy Process (FAHP) and the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS), to design and evaluate home fitness equipment that aligns with contemporary demands. First, we used crawler data to collect user reviews of home fitness equipment from a well-known Chinese shopping platform. The data were cleaned and processed to extract key user needs and preferences. Next, the FAHP method was used to prioritize these requirements, and TOPSIS was applied for the comprehensive evaluation of design proposals. This process allowed us to identify the solution that best meets user needs, completing the development of the product design. The results indicate that the second design, with its features targeting lumbar health, efficient space utilization, rich interactive experience, integration of smart technology, and minimalist appearance, has significant market potential and social value. Finally, the SUS (System Usability Scale) was used to validate the design, showing excellent user satisfaction and usability for the second scheme. This study establishes a design process incorporating web scraping, FAHP, and TOPSIS, demonstrating the effectiveness of this theoretical integration in the field of home fitness equipment design.

Author 1: Jinyang Xu

Author 2: Xuedong Zhang

Author 3: Xinlian Li

Author 4: Shun Yu

Author 5: Yanming Chen

Keywords: Home fitness equipment; crawler data; FAHP; TOPSIS; product design

Paper 43: Evaluating the Impact of Fuzzy Logic Controllers on the Efficiency of FCCUs: Simulation-Based Analysis

Abstract: This study investigates the methods for creating nonlinear models and developing Fuzzy logic controllers for the Fluidized Catalytic Cracking Unit (FCCU) at different global refineries. The FCCU plays a crucial role in the petrochemical sector, processing a significant portion of the world's crude oil - in 2006, FCCUs were responsible for refining a third of the global crude oil supply. These units are essential for converting heavier oils, such as gasoil and crude oil, into lighter, more critical products like gasoline and olefinic gases. Given their efficiency in producing a large volume of products and the volatile nature of petrochemical market prices, optimization of these units is a priority for engineers and investors. Traditional control mechanisms often need to improve in managing the FCCU's complex, dynamic, and nonlinear operations, where creating an accurate mathematical model is challenging or involves significant simplifications. Fuzzy Logic controllers, which mimic human reasoning more closely than conventional methods, offer a promising alternative for such unpredictable and complex systems. The results of this work illustrate the usefulness and possible advantages of utilizing Fuzzy Logic controllers in the management of FCCU plants and they are also compared with the latest machine learning techniques as well. These findings are corroborated by simulations conducted with the MATLAB Fuzzy Logic Toolbox R2012b.

Author 1: Harsh Pagare

Author 2: Kushagra Mishra

Author 3: Kanhaiya Sharma

Author 4: Sandeep Singh Rawat

Author 5: Shailaja Salagrama

Keywords: Non-Linear modeling; fuzzy logic controller; machine learning; optimization

Paper 44: Hybrid Machine Learning Approach for Real-Time Malicious URL Detection Using SOM-RMO and RBFN with Tabu Search

Abstract: The proliferation of malicious URLs has become a significant threat to internet security, encompassing SPAM, phishing, malware, and defacement attacks. Traditional detection methods struggle to keep pace with the evolving nature of these threats. Detecting malicious URLs in real-time requires advanced techniques capable of handling large datasets and identifying novel attack patterns. The challenge lies in developing a robust model that combines efficient feature extraction with accurate classification. We propose a hybrid machine learning approach combining Self-Organizing Map based Radial Movement Optimization (SOM-RMO) for feature extraction and Ensemble Radial Basis Function Network (RBFN) based Tabu Search for classification. SOM-RMO effectively reduces dimensionality and highlights significant features, while RBFN, optimized with Tabu Search, classifies URLs with high precision. The proposed model demonstrates superior performance in detecting various malicious URL attacks. On a benchmark dataset, the proposed approach achieved an accuracy of 96.5%, precision of 95.2%, recall of 94.8%, and an F1-score of 95.0%, outperforming traditional methods significantly.

Author 1: Swetha T

Author 2: Seshaiah M

Author 3: Hemalatha K L

Author 4: Murthy S V N

Author 5: Manjunatha Kumar BH

Keywords: Malicious URL detection; self-organizing map; Radial Movement Optimization; ensemble radial basis function network; Tabu Search

Paper 45: Missing Value Imputation in Data MCAR for Classification of Type 2 Diabetes Mellitus and its Complications

Abstract: Type 2 diabetes mellitus (T2DM) is a disease that is at risk for many complications. Previous research on the prognosis of T2DM and its complications is limited to the impact of T2DM on one particular disease. Guidebook for T2DM Management in Indonesia has eight categories of T2DM complications. The purpose of this study is to classify T2DM prognosis into eight categories: one controlled class and seven classes of aggravating disorders. The classification was based on medical record data from T2DM patients at Panti Rapih Hospital in Yogyakarta between 2017 and 2022. The problem is that the medical record data has numerous missing values (MV). The dataset had 29% missing values, classified as Missing Completely at Random (MCAR). This study performed imputation on the dataset prior to categorization. For MV imputation, a variety of imputation methods were used, and their accuracy was measured using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The best imputation results were utilized to update the dataset. Subsequently, the dataset was used for classification employing several classification methods. The classification results were compared to determine the method with the highest accuracy in this scenario. The Decision Tree method with stratified k-fold cross-validation emerged as the optimal method for this classification. The results revealed an average accuracy value of 0.8529.

Author 1: Anik Andriani

Author 2: Sri Hartati

Author 3: Afiahayati

Author 4: Cornelia Wahyu Danawati

Keywords: Missing value; prognosis of diabetes mellitus; missing completely at random; decision tree

Paper 46: Optimizing Dance Training Programs Using Deep Learning: Exploring Motion Feedback Mechanisms Based on Pose Recognition and Prediction

Abstract: Dance pose recognition and prediction is an important part of dance training and a challenging task in the field of artificial intelligence. Due to the diverse styles and significant variations in dance movements, conventional methods struggle to capture effective dance pose features for recognition. In this context, we have developed a dance pose recognition and prediction method based on deep learning. Given the characteristics of dance movements, such as complex human postures and dynamic movements, we proposed the MKFF-ST-GCN model, which integrates multi-kinematic feature fusion with ST-GCN. This model fully captures the dynamic information of dance movements by calculating the first and second-order kinematic features of keypoints and fuses the kinematic features using a multi-head attention mechanism. Additionally, to address dance pose prediction issues, we proposed the STGA-Net based on the spatial-temporal graph attention mechanism. This model improves the long-distance information modeling capability by calculating local and global graph attentions of dance poses, effectively solving the problem of dance pose prediction. To comprehensively evaluate the quality of the proposed methods in dance pose recognition and prediction, we conducted extensive experimental validations and comparisons with several common algorithms. The experimental results fully demonstrate the effectiveness of our methods in dance pose recognition and prediction. This study not only advances the technology of dance pose recognition and prediction but also provides valuable experience for the field.

Author 1: Yuting Jiao

Keywords: Deep learning; pose recognition; pose prediction; dance training; graph convolutional network; attention mechanism

Paper 47: Improved Decision Support System for Alzheimer's Diagnosis Using a Hybrid Machine Learning Approach with Structural MRI Brain Scans

Abstract: Alzheimer's disease (AD) causes damage to brain cells and their activities. This disease is typically caused by ageing, making people over the age of 65 more susceptible. As the disease progresses, it slowly destroys brain cells, making it harder to think clearly, recall things, and do everyday tasks. The end result of this is dementia. Metabolic disorders, such as diabetes and Alzheimer's disease, affect a substantial proportion of the world's population. While there is no permanent cure for AD, early diagnosis can help reduce damage to brain cells and support a faster recovery. Recent research has explored various machine learning approaches for early disease detection. However, traditional ML (Machine Learning) methods and deep learning techniques such as CNNs have not been individually effective in accurately detecting Alzheimer's disease (AD). In this proposed work, we developed a hybrid model that processes sMRI brain images to detect them as demented or non-demented. The model consists of two parts: the first part involves extracting significant features through a sequence of convolution and pooling operations; the second part uses these features to train SVM for binary classification. Data augmentation techniques such as horizontal flipping are used to balance dataset. We calculated key performance metrics essential for the healthcare domain, including sensitivity, specificity, accuracy, and F1-score. Notably, our model achieved an impressive accuracy of 99.60% in detecting AD, with a sensitivity of 99.83%, a specificity of 99.40%, and an F1-score of 99.58%. These results were validated using 15-fold cross-validation, enhancing the model's robustness for new data. This approach yields a more robust model, offering greater accuracy and precision compared to existing methods. This model can effectively support manual systems for detecting AD with greater accuracy.

Author 1: Niranjan Kumar Parvatham

Author 2: Lakshmana Phaneendra Maguluri

Keywords: Alzheimer’s disease; binary classification; Convolutional Neural Network (CNN); horizontal flipping; healthcare decision support system; MRI images; Support Vector Machine(SVM)

Paper 48: Innovative Melanoma Diagnosis: Harnessing VI Transformer Architecture

Abstract: Melanoma, the most severe type of skin cancer, ranks ninth among the most prevalent cancer types. Prolonged exposure to ultraviolet radiation triggers mutations in melanocytes, the pigment -producing cells responsible for melanin production. This excessive melanin secretion leads to the formation of dark-colored moles, which can evolve into cancerous tumors over time and metastasize rapidly. This research introduces a Vision Transformer, revolutionizes computer vision architecture by diverging from traditional convolutional neural networks, employing transformer models to handle images as sequences of flattened, spatially-structured patches. The dermoscopy images sourced from the Kaggle repository, an extensive online database known for its diverse collection of high-quality medical imagery is utilized in this study. This novel deep learning model for melanoma classification, aiming to enhance diagnostic accuracy and reduce reliance on expert interpretation. The model achieves an accuracy of 96.23%, indicating strong overall correctness in classifying both Benign and Malignant cases. Comparative simulation of the proposed method against other methods in skin cancer diagnosis reveal that the suggested approach attains superior accuracy. These findings underscore the efficacy of the system in advancing the field of skin cancer diagnosis, offering promising prospects for enhanced accuracy and efficacy in clinical settings.

Author 1: Sreelakshmi Jayasankar

Author 2: T. Brindha

Keywords: Vision transformer; melanoma; convolutional neural networks; deep learning model; transformer encoder; dermoscopy image

Paper 49: Enhancing Safety for High Ceiling Emergency Light Monitoring

Abstract: The performance of information technology has gradually improved and advanced during this period for safety management. Nonetheless, there is no disputing that during power outages, emergency lights continue to be crucial to people's safety and comfort. Regrettably, businesses don't give emergency lighting in buildings enough thought. The primary causes of the problem are expensive spending and long-term management for high ceiling safety. Additionally, one aspect that must be considered and ensured is the safety of maintenance personnel when the light is installed in high-rise locations. Thus, by creating wireless global control and monitoring via Android mobile phones, our effort intends to increase the availability and reliability of the emergency light. The suggested light monitoring system collects information from Internet of Things devices and transmits it to users' mobile phones over the Internet. Moreover, the risk of employees keeping the emergency light will be significantly reduced because it is monitored via the Internet on mobile devices. Additionally, by using the information the sensor inside the emergency light collects, it is possible to estimate its current condition, including its battery life. This repair will also improve everyone's safety within the building by increasing the emergency light's dependability with good process innovation.

Author 1: G. X. Jun

Author 2: M. Batumalay

Author 3: C. Batumalai

Author 4: Prabadevi B

Keywords: Safety management; Internet of Things; high ceiling safety; high building safety; safety; process innovation

Paper 50: Data-Driven Approaches to Energy Utilization Efficiency Enhancement in Intelligent Logistics

Abstract: With the rapid development of intelligent logistics, new challenges and opportunities are presented for energy utilization efficiency improvement. This study explores the feasibility and effectiveness of using data-driven methods to improve energy utilization efficiency in an intelligent logistics environment and provides theoretical support and practical guidance for achieving the sustainable development of optimized logistics management procedures. First, a dataset was established by collecting relevant data in the optimized logistics management procedure, including transportation information and energy consumption data. Then, data analysis and mining techniques are used to conduct an in-depth dataset analysis to reveal the influencing factors of energy utilization efficiency and potential optimization directions. Then, strategies and methods for energy utilization efficiency improvement are designed by combining intelligent optimization algorithms. Finally, simulation experiments and case studies are utilized to verify the effectiveness and feasibility of the proposed methods. The results show that using data-driven methods can significantly improve the energy utilization efficiency of optimized logistics management procedures, reduce logistics costs, and enhance the sustainability and competitiveness of the system. Through in-depth analysis and empirical research, a series of actionable optimization strategies are proposed, providing new ideas and methods for optimizing energy and logistics management procedures. These results significantly promote the sustainable development of optimized logistics management procedures and enhance competitiveness.

Author 1: Xuan Long

Keywords: Intelligent logistics; energy; utilization efficiency; data-driven

Paper 51: Design and Application of the DPC-K-Means Clustering Algorithm for Evaluation of English Teaching Proficiency

Abstract: Effective and precise methodologies for evaluating the proficiency in English language instruction are instrumental in enhancing educators' competencies and the effectiveness of educational administrative processes. The objective of this paper is to refine the neutrality and precision of such assessments by introducing a novel approach that leverages an advanced K-means algorithm in conjunction with convolutional neural networks (CNNs). Initially, a thorough examination of the issue at hand leads to the formulation of an assessment framework that integrates both a clustering algorithm and a CNN, with a comprehensive elucidation of the pivotal technical aspects. Subsequently, the paper introduces a data clustering and categorization technique grounded in the DPC-K-means methodology, specifically tailored for indices that measure English teaching proficiency, and employs CNNs to devise a model for evaluating these competencies. The integration of these two components—data clustering and the assessment model—gives rise to an innovative technique. Ultimately, the proposed method is implemented and its practicality is substantiated through an analysis of empirical data from educators' teaching proficiency indices. A comparative analysis with existing algorithms reveals that the proposed method achieves superior clustering performance and the lowest margin of error in predictive assessments.

Author 1: Mei Niu

Keywords: K-Means; density-peak clustering algorithm; ELT competency assessment; convolutional neural network

Paper 52: Enhancing Indonesian Text Summarization with Latent Dirichlet Allocation and Maximum Marginal Relevance

Abstract: Maximum Marginal Relevance (MMR) Summarization of text is very important in grasping quickly long articles particularly for people who are very busy. In this paper, we use LDA to give topic queries for news articles, which then become inputs to the MMR method. According to this paper's summarization system, the ROUGE metric is employed to evaluate the summaries of news articles with 30 percent compression and 50 percent compression. Experimental findings show that the LDA-MMR combination outperforms MMR on its own in all our tests across all query lengths or number of sentences used and gives highest average ROUGE value of 0.570 for a 50% compression rate; 0.547 at 30% This implies that our system efficiently produces meaningful summaries using content-based keywords rather than click bait titles, which should not lead to complaints about misleading advertisements. This summarizer can convey the main points of a piece of news coverage in a concise form, thus offering people useful new tools for quickly digesting information.

Author 1: Muhammad Faisal

Author 2: Bima Hamdani Mawaridi

Author 3: Ashri Shabrina Afrah

Author 4: Supriyono

Author 5: Yunifa Miftachul Arif

Author 6: Abdul Aziz

Author 7: Linda Wijayanti

Author 8: Melisa Mulyadi

Keywords: Indonesian summarization; LDA; MMR; ROUGE evaluation

Paper 53: Research on Traffic Flow Prediction Using the MSTA-GNet Model Based on the PeMS Dataset

Abstract: This study introduces the MSTA-GNet (Multi-Scale Spatiotemporal Attention Graph Network), a novel deep learning model which integrates spatiotemporal self-attention mechanisms to model heterogeneous dependencies in traffic networks. The primary objective of the study is to improve existing traffic flow prediction models to address the inadequacies of traditional models in complex big data environments. Key innovations of the MSTA-GNet model include positional encoding and global and local self-attention mechanisms to capture long-term and short-term dependencies. Using the PeMS (Performance Measurement System) dataset, the study conducted performance comparison experiments among various deep learning models, including LSTM (Long Short-Term Memory), GCN (Graph Convolutional Network), DCRNN (Diffusion Convolutional Recurrent Neural Network), STGCN (Spatiotemporal Graph Convolutional Network), STMetaNet (Spatiotemporal Meta Network), and MSTA-GNet. The results showed that MSTA-GNet significantly outperformed other models with improvements of 13.4%, 11.8%, and 9.7% in Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) metrics, respectively. Ablation studies further validated the significance of attention mechanisms, feature extraction, convolutional layers, and graph networks, confirming the effectiveness and practical application of MSTA-GNet in traffic flow prediction. This research provides important insights for AI-based congestion management, support for low-carbon traffic networks, and optimization of local traffic operations, demonstrating its significant practical value in intelligent transportation systems.

Author 1: Deng Cong

Keywords: MSTA-GNet; deep learning; PeMS dataset; traffic flow prediction

Paper 54: Laboratory Abnormal Behavior Recognition Method Based on Skeletal Features

Abstract: The identification of abnormal laboratory behavior is of great significance for the safety monitoring and management of laboratories. Traditional identification methods usually rely on cameras and other equipment, which are costly and prone to privacy leakage. In the process of human body recognition, they are easily affected by various factors such as complex backgrounds, human clothing, and light intensity, resulting in low recognition rates and poor recognition results. This article investigates a laboratory abnormal behavior recognition method based on skeletal features. One is to use Kinect sensors instead of traditional image sensors to obtain characteristic skeletal data of the human body, reducing external limitations such as lighting and increasing effective data collection. Then, the collected data is smoothed, aligned, and image enhanced using moving average filtering, discrete Fourier transform, and contrast, effectively improving data quality and helping to better identify abnormal behavior. Finally, the OpenPose algorithm is used to construct a laboratory anomaly behavior recognition model. OpenPose can be used to connect the entire skeleton through the relationships between points during the process of extracting human skeletal points, and combined with multi-scale pyramid networks to improve the network structure, effectively improving the accuracy and recognition speed of laboratory abnormal behavior recognition. The experiment shows that the accuracy, precision, and recall of the behavior recognition model constructed by the algorithm are 95.33%, 96.68%, and 93.77%, respectively. Compared with traditional anomaly detection methods, it has higher accuracy and robustness, lower parameter count, and higher operational efficiency.

Author 1: Dawei Zhang

Keywords: Skeletal features; abnormal behavior recognition; OpenPose algorithm; Kinect sensor; Discrete Fourier Transform

Paper 55: Twitter Truth: Advanced Multi-Model Embedding for Fake News Detection

Abstract: The identification of fake news represents a substantial challenge within the context of the accelerated dissemination of digital information, most notably on social media and online platforms. This study introduces a novel approach, entitled " MT-FND: Multi-Model Embedding Approach to Fake News Detection," which is designed to enhance the detection of fake news. The methodology presented here integrates the strengths of multiple transformer-based models, namely BERT, ELECTRA, and XLNet, with the objective of encoding and extracting contextual information from news articles. In addition to transformer embeddings, a variety of other features are incorporated, including sentiment analysis, tweet length, word count, and graph-based features, to enrich the representation of textual content. The fusion of signals from diverse models and features provides a more comprehensive and nuanced comprehension of news articles, thereby improving the accuracy of discerning misinformation. To evaluate the efficacy of the approach, a benchmark dataset comprising both authentic and fabricated news articles was employed. The proposed framework was tested using three different machine-learning models: Random Forest (RF), Support Vector Machine (SVM), and XGBoost (XGB). The experimental results demonstrate the effectiveness of the multi-model embedding fusion approach in detecting fake news, with XGB achieving the highest performance with an accuracy of 87.28%, a precision of 85.56%, a recall of 89.53%, and an F1-score of 87.50%. These findings signify a notable improvement over traditional machine learning classifiers, underscoring the potential of this fusion approach in advancing methodologies for combating misinformation, promoting information integrity, and enhancing decision-making processes in digital media landscapes.

Author 1: Yasmine LAHLOU

Author 2: Sanaa El FKIHI

Author 3: Rdouan FAIZI

Keywords: Fake news detection; transformer-based models; text classification; sentiment analysis

Paper 56: Protein-Coding sORFs Prediction Based on U-Net and Coordinate Attention with Hybrid Encoding

Abstract: Small proteins encoded by small open reading frames (sORFs) exhibit significant biological activity in crucial biological processes such as embryonic development and metabolism. Accurately predicting whether sORFs encode small proteins is a key challenge in current research. To address this challenge, many methods have been proposed, however, existing methods rely solely on biological features as the sequence encoding scheme, which results in high feature extraction complexity and limited applicability across species. To tackle this issue, we proposed a deep learning architecture UAsORFs based on hybrid coding of sORFs sequences. In contrast to mainstream prediction methods, this framework processes sORF sequences using a mixed encoding approach, including both one-hot and gapped k-mer encodings, which effectively captures global and local sequence information. Additionally, it autonomously learns to extract features of sORFs and captures both long-range and short-range interactions between sequences through U-Net and coordinate attention mechanisms. Our research demonstrates significant progress in predicting encoded peptides from eukaryotic and prokaryotic sORFs, particularly in improving the cross-species predictive MCC index on the eukaryotic dataset.

Author 1: Ziling Wang

Author 2: Wenxi Yang

Author 3: Zhijian Qu

Keywords: Small open reading frames; deep learning; hybrid coding; U-Net; coordinate attention

Paper 57: An Improved YOLOv8 Method for Measuring the Body Size of Xinjiang Bactrian Camels

Abstract: Camel body size measurement has initially been applied in livestock production. However, current methods suffer from low measurement accuracy due to detection box localization loss and occlusions. This study proposes an effective algorithm, Camel-YOLOv8, specifically designed for detecting Xinjiang Bactrian camels and calculating their body sizes. By integrating the Selective Kernel Networks (SKAttention) mechanism and an enhanced Asymptotic Feature Pyramid Network structure (AFPN-beta), the algorithm successfully captures the body characteristics of Bactrian camels in natural environments and converts these into precise size data. We have developed a Xinjiang Bactrian camel body size measurement dataset and applied the enhanced YOLOv8 model for accurate classification and detection. By extracting key point pixel values and applying Zhang Zhengyou’s calibration method, the coordinate system data is converted into accurate body size measurements. The Camel-YOLOv8 achieves a detection accuracy of 76.4% on the Xinjiang Bactrian camel dataset, marking a 3.7% improvement over the baseline model. In terms of body size calculation, the average relative errors for height and chest circumference are -3.39% and 4.1%, respectively, demonstrating high measurement precision. The algorithm not only maintains high detection accuracy but also achieves a reasonable balance between detection speed and efficiency, providing an effective solution for rapid acquisition of camel body size information.

Author 1: Yue Peng

Author 2: Alifu Kurban

Author 3: Mengmei Sang

Keywords: YOLOv8; Asymptotic Feature Pyramid Network; SKAttention; Bactrian camel body size measurement

Paper 58: Towards Secure Internet of Things-Enabled Healthcare: Integrating Elliptic Curve Digital Signatures and Rivest Cipher Encryption

Abstract: The expansion of Internet of Things (IoT) applications, such as wireless sensor networks, intelligent devices, Internet technologies, and machine-to-machine interaction, has changed current information technology in recent decades. The IoT enables the exchange of information and communication between items via an internal network. Nevertheless, the advancement of technology raises the urgent issue of ensuring data privacy and security, particularly in critical sectors like healthcare. This study aims to address the problem by developing a hybrid security scheme that combines the Secure Hash Algorithm (SHA-256), Rivest Cipher 4 (RC4), and Elliptic Curve Digital Signature Scheme (ECDSS) to ensure the confidentiality and integrity of medical data transmitted by IoT-enabled healthcare systems. This hybrid model employs the Elliptic Curve Digital Signature Scheme (ECDSS) to perform exclusive OR (XOR) operations inside the RC4 encryption algorithm. This enhances the RC4 encryption process by manipulating the encryption key. Moreover, SHA-256 is used to convert incoming data in order to guarantee data security. An empirical investigation validates the superiority of the suggested model. This framework attains a data transfer rate of 11.67 megabytes per millisecond, accompanied by an encryption duration of 846 milliseconds and a decryption duration of 627 milliseconds.

Author 1: Longyang Du

Author 2: Tian Xie

Keywords: IoT-enabled healthcare; data privacy; security; hybrid security framework; SHA-256; RC4; encryption; data integrity

Paper 59: Dose Archiving and Communication System in Moroccan Healthcare: A Unified Approach to X-Ray Dose Management and Analysis

Abstract: This study explores the implementation of a Dose Archiving and Communication System (DACS) in Moroccan healthcare, highlighting the importance of X-ray dose management in modern radiology. It emphasizes patient safety and the ALARA principle to minimize radiation exposure while maintaining diagnostic accuracy. The research discusses advancements in imaging technologies, such as dose-reduction algorithms and real-time monitoring systems. A survey of 1000 healthcare professionals reveals significant challenges in X-ray dose management, including poor dose tracking, regulatory non-compliance, and inadequate radiation protection training. Noteworthy findings reveal that 10% of patients received doses exceeding 5 Gray, underscoring the exigency for robust dose management systems. The article delineates a strategic implementation approach for DACS in Moroccan hospitals, comprising meticulous needs assessment, infrastructure fortification, and stakeholder engagement. By harnessing cloud-based storage, blockchain technology, and industry-standard encryption protocols, the envisioned DACS endeavors to furnish a secure, scalable, and efficient framework for radiation dose management. This holistic approach, underpinned by empirical statistics regarding training in radiation protection, network infrastructure, and DACS implementation strategies, aims to elevate patient outcomes and ensure stringent regulatory compliance.

Author 1: Lhoucine Ben Youssef

Author 2: Abdelmajid Bybi

Author 3: Hilal Drissi

Author 4: El Ayachi Chater

Keywords: DACS; Real-time monitoring; radiation protection; radiology practice; healthcare professionals; x-ray doses; regulatory compliance; patient safety; Moroccan healthcare

Paper 60: UTAUT Model for Digital Mental Health Interventions: Factors Influencing User Adoption

Abstract: The impact of digital revolution on mental health therapies is examined in this research. As explained in the paper, the delivery of mental healthcare is being revolutionized by digital transformation, which is providing creative answers to the problems associated with mental health illnesses. But knowing and managing user approval is crucial to the effective integration of digital transformation approaches into mental health therapies. To investigate user’s acceptance regarding digital transformation in Mental Health therapies, this study outlines a modeling-based method based on a well-established Unified Theory of acceptability and Use of Technology theory, abbreviated as UTAUT. This study delves into the base constructs of Expected Performance, Expected Effort, Social Influence, Conditions facilitating the use of proposed solution, Hedonic Motivation, and Value for Money, utilizing the UTAUT model as a framework. By employing Structural Equation Modelling (SEM) in a thorough study, the goal of this research is to identify statistical correlations that impact user acceptance dynamics. To offer context-specific insights, this article also delves into digital mental health solutions, including teletherapy platforms, mood monitoring smartphone applications, and virtual reality-based exposure therapy. This study enhances accessibility, engagement, and results for people seeking mental health care by providing a deeper knowledge of user acceptability, which aids in the creation and roll-out of digital mental health solutions.

Author 1: Mohammed Alojail

Keywords: Digital transformation; mental health interventions; UTAUT model; TAM; SmartPLS; user acceptance; hypothesis testing

Paper 61: ResNet50 and GRU: A Synergistic Model for Accurate Facial Emotion Recognition

Abstract: Humans use voice, gestures, and emotions to communicate with one another. It improves oral communication effectiveness and facilitates concept of understanding. Majority of people are able to identify facial emotions with ease, regardless of gender, nationality, culture, or ethnicity. The recognition of facial expressions is becoming more and more significant in a variety of newly developed computing applications. Facial expression detection is a hot topic in almost every industry, including marketing, artificial intelligence, gaming, and healthcare. This study proposes a novel hybrid model combining ResNet-50 and Gated Recurrent Unit (GRU) for enhanced Facial emotion recognition (FER) accuracy. The dataset for the study is taken from Kaggle repository. ResNet-50, a deep convolutional neural network, excels in feature extraction by capturing intricate spatial hierarchies in facial images. GRU, effectively processes sequential data, capturing temporal dependencies crucial for emotion recognition. The integration of ResNet-50 and GRU leverages the strengths of both architectures, enabling robust and accurate emotion detection. Experimental result on CK+ dataset demonstrate that the proposed hybrid model outperforms current methods, achieving a remarkable accuracy of 95.56%. This superior performance underscores the model's potential for real-world applications in diverse domains such as security, healthcare, and interactive systems.

Author 1: Shanimol. A

Author 2: J Charles

Keywords: Deep convolutional neural network; ResNet-50; Facial Emotion Recognition; Gated Recurrent Unit

Paper 62: Efficient Parallel Algorithm for Extracting Fuzzy-Crisp Formal Concepts

Abstract: Fuzzy Formal Concept Analysis (FFCA) is a robust mathematical tool for analyzing data, particularly where uncertainty or fuzziness is inherent. FFCA is utilized across various domains, including data mining, information retrieval, and knowledge representation. However, fuzzy concepts extraction is a crucial yet computationally intensive task. This paper addresses the challenge of time efficiency in extracting single-sided fuzzy concepts from large datasets. A parallel algorithm is proposed to reduce computational time and optimize resource utilization, thus enabling the scalable analysis of expanding datasets. By computing fuzzy concepts across multiple threads in parallel, each thread processes an attribute independently to extract fuzzy concepts, which are then merged in the final step. The proposed algorithm extracts fuzzy-crisp concepts, which are more concise than other types of fuzzy concepts. Experiments were conducted to evaluate the performance of the proposed parallel algorithm against existing sequential methods. Experimental results demonstrate significant gains in computational efficiency, with the algorithm achieving an average time reduction of 68% compared to the attribute-based algorithm and up to 83%-time reduction compared to the fuzzy CbO algorithm across various types of datasets, including binary, quantitative, and fuzzy.

Author 1: Ebtesam Shemis

Author 2: Arabi Keshk

Author 3: Ammar Mohammed

Author 4: Gamal Elhady

Keywords: Fuzzy Formal Concept Analysis; single-sided fuzzy concept; fuzzy-crisp concepts; parallel algorithm; fuzzy concepts extraction; knowledge representation

Paper 63: Automatic Plant Disease Detection System Using Advanced Convolutional Neural Network-Based Algorithm

Abstract: With technology innovations such as Artificial Intelligence (AI) and Internet of Things (IoT), unprecedented applications and solutions to real world problems are made possible. Precision agriculture is one such problem which is aimed at technology driven agriculture. So far, the research on agriculture and usage of technologies are at government level to reap benefits of technologies in crop yield prediction and finding the cultivated areas. However, the fruits of technologies could not reach farmers. Farmers still suffer from plenty of problems such as natural calamities, reduction in crop yield, high expenditure and lack of technical support. Plant diseases is an important problem faced by farmers as they could not find diseases early. There is need for early plant disease detection in agriculture. From the related works, it is known that deep learning techniques like Convolutional Neural Network (CNN) is best used to process image data to solve real world problems. However, as one size does not fit all, CNN cannot solve all problems without exploiting its layers based on the problem in hand. Towards this end, we designed an automatic plant disease detection system by proposing an advanced CNN model. We proposed an algorithm known as Advanced CNN for Plant Disease Detection (ACNN-PDD) to realize the proposed system. Our system is evaluated with PlantVillage, a benchmark dataset for crop disease detection result, and real-time dataset (captured from live agricultural fields). The investigational outcomes showed the utility of the proposed system. The proposed advanced CNN based model ACNN-PDD achieve 96.83% accuracy which is higher than many existing models. Thus our system can be integrated with precision agriculture infrastructure to enable farmers to detect plant diseases early.

Author 1: Sai Krishna Gudepu

Author 2: Vijay Kumar Burugari

Keywords: Plant disease detection; advanced CNN; Artificial Intelligence (AI); deep learning; precision agriculture

Paper 64: Towards Secure Cloud-Enabled Wireless Ad-Hoc Networks: A Novel Cross-Layer Validation Mechanism

Abstract: Network security tackles a broad spectrum of damaging activities that threaten network infrastructure. Addressing these risks is essential to keep data accurate and networks running. This research aims to detect and prevent blackholes and wormholes in cloud-based wireless ad-hoc networks. A new Cross-Layer Validation Mechanism (CLVM) is introduced to detect and counter these dangerous attacks. CLVM boosts network security and ensures data travels through cross-layer interactions. CLVM is tested using NS2 software by performing several simulations and comparing the results with previous methods. The results show that CLVM effectively defends against blackhole and wormhole attacks, which makes it a crucial extra service for cloud computing. CLVM provides a strong defense against new security threats, making sure the network stays reliable and safe.

Author 1: Zhenguo LIU

Keywords: Network security; wireless ad-hoc networks; cloud environments; cross-layer validation; blackhole and wormhole attacks

Paper 65: UWB Printed MIMO Antennas for Satellite Sensing System (SRSS) Applications

Abstract: The deployment of ultra-wideband (UWB) technology offers enhanced capabilities for various Internet of Things (IoT) applications, including smart cities, smart buildings, smart aggregation, and smart healthcare. UWB technology supports high data rate communication over short distances with very low power densities. This paper presents a UWB printed antenna design with multiple input and output (MIMO) capabilities, specifically tailored for Routed Satellite Sensor Systems (SRSS) to enhance IoT applications. The proposed UWB printed antenna, designed for the 2–18 GHz frequency band, has overall dimensions of 14.5 mm x 14.5 mm, with an efficiency exceeding 70% and a gain ranging from 2 to 6.5 dB. Both simulated and measured reflection parameters (|S11|) at the antenna input show strong agreement. Furthermore, a compact MIMO system is introduced, featuring four closely spaced antennas with a gap of 0.03λ, housed in a 60 mm x 48 mm module. To minimize coupling effects between the antennas, the design incorporates five Split Ring Resonator (SRR) elements arranged linearly between the radiating elements. This arrangement achieves a mutual coupling reduction to -35 dB at 8 GHz, compared to -20 dB isolation in systems without SRR. The results demonstrate that the proposed MIMO antenna system offers promising performance and meets the requirements for effective space communication within satellite sensor networks.

Author 1: Wyssem Fathallah

Author 2: Chafai Abdelhamid

Author 3: Chokri Baccouch

Author 4: Alsharef Mohammad

Author 5: Khalil Jouili

Author 6: Hedi Sakli

Keywords: 5G antenna; 5G satellite networks; millimeter band; wireless communications; SRR; IoT

Paper 66: Novel Data-Driven Machine Learning Models for Heating Load Prediction: Single and Optimized Naive Bayes

Abstract: Numerous approaches can be employed to create models for assessing the heat gains of a building arising from both external and internal sources. This modeling process evaluates effective operational strategies, conducts retrofit audits, and projects energy consumption. These techniques range from simple regression analyses to more intricate models grounded in physical principles. A prevalent assumption underlying all these modeling techniques is the requirement for input variables to be derived from authentic data, as the absence of realistic input data can lead to substantial underestimations or overestimations in energy consumption assessments. In this paper, eight input parameters, including relative compactness, orientation, wall area, roof area, glazing area, overall height, surface area, and glazing area distribution, are employed for training proposed Naive Bayes (NB)-based machine learning models. Utilizing a novel approach, this research explores the application of Beluga Whale Optimization and the Coot Optimization algorithm for optimizing the Naive Bayes model in heating load prediction. By harnessing the collective intelligence of Beluga Whales and drawing from the cooperative behavior of coots, the research aims to improve the model's predictive capabilities, which is of paramount importance in building energy management. Based on the comparative analysis between developed models (NB, NBCO, and NBBW), it is attainable that NBCO and NBBW, as two optimized models, have 2.4% and 1.3% higher R2 values, respectively. Also, the RMSE of the NBCO was, on average, 19-33% lower than that of the two other models, confirming the high accuracy of NBCO. This innovative integration of bio-inspired optimization techniques with machine learning demonstrates a promising avenue for optimizing predictive models, offering potential energy efficiency and sustainability advancements in the built environment.

Author 1: Fangyuan Li

Keywords: Prediction models; heating load demand; building energy consumption; Naive Bayes; metaheuristic optimization algorithms

Paper 67: Facial Expression Real-Time Animation Simulation Technology for General Mobile Platforms Based on OpenGL

Abstract: With the popularization and development of mobile devices, the demand for image processing continues to increase. However, the limited hardware resources of mobile devices make traditional CPU computing unable to meet the requirements of real-time image processing. In response to the limited rendering resources of mobile platforms, this study adopts OpenGL for graphic interface design and animation simulation of facial expressions to control the changes of facial expressions in real-time, achieve facial expression animation simulation, and develop effective expression fusion methods. By combining rich rendering effects such as particle effects, facial expressions can be expressed more realistically and interestingly in 3D models. The results indicated that the research method only required less than 50 MB of memory, and the average accuracy of facial expression recognition had significantly improved. The final normalized average error level was close to 4%, with higher accuracy. The processing speed of each image was around 19.4ms, which could achieve animation simulation of facial expressions and had strong universality and flexibility. This method optimizes the real-time performance, stability, and user experience of facial expression real-time animation simulation, which can meet the needs of different application scenarios.

Author 1: Mingzhe Cao

Keywords: OpenGL; mobile platform; facial expressions; animation simulation; rendering

Paper 68: Noise Reduction Techniques in Adas Sensor Data Management: Methods and Comparative Analysis

Abstract: This review examines noise reduction techniques in Advanced Driver Assistance Systems (ADAS) sensor data management, crucial for enhancing vehicle safety and performance. ADAS relies on real-time data from conventional sensors (e.g., wheel speed sensors, LiDAR, radar, cameras) and MEMS sensors (e.g., accelerometers, gyroscopes) to execute critical functions like lane keeping, collision avoidance, and adaptive cruise control. These sensors are susceptible to thermal noise, mechanical vibrations, and environmental interferences, which degrade system performance. We explore filtering techniques including KalmanNet, Simple Moving Average (SMA), Exponential Moving Average (EMA), Wavelet Denoising, and Low Pass Filtering (LPF), assessing their efficacy in noise reduction and data integrity improvement. These methods are compared using key performance metrics such as Signal-to-Noise Ratio (SNR), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). Recent advancements in hybrid filtering approaches and adaptive algorithms are discussed, highlighting their strengths and limitations for different sensor types and ADAS functionalities. Findings demonstrate the superior performance of Wavelet Denoising for non-stationary signals, SMA and EMA's effectiveness for smoother signal variations, and LPF's excellence in high-frequency noise attenuation with careful tuning. KalmanNet showed significant improvements in noise reduction and data accuracy, particularly in complex and dynamic environments. Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) were especially effective for RADAR sensors, handling non-linearities and providing accurate state estimation. Emphasizing Hardware-in-the-Loop (HIL) bench testing to validate these techniques in real-world scenarios, this study underscores the importance of selecting appropriate methods based on specific noise characteristics and system requirements. This research provides valuable insights for ADAS and autonomous driving technologies development, emphasizing precise signal processing's critical role in ensuring accurate sensor data interpretation and decision-making.

Author 1: Ahmed Alami

Author 2: Fouad Belmajdoub

Keywords: ADAS; sensor data management; noise reduction; KalmanNet; Wavelet Denoising; RADAR; SMA; EMA; LPF; HIL bench testing

Paper 69: DMMFnet: A Dual-Branch Multimodal Medical Image Fusion Network Using Super Token and Channel-Spatial Attention

Abstract: Multimodal medical image fusion leverages the correlation between different modal images to enhance the information contained within a single medical image. Existing fusion methods often fail to effectively extract multiscale features from medical images and establish long-distance relationships between deep feature blocks. To address these issues, we propose DMMFnet, an encoder-decoder fusion network that utilizes shared and private encoders to extract shared and private features. DMMFnet is based on super token sampling and channel-spatial attention. The shared encoder and decoder use a transformer structure with super token sampling technology to effectively integrate information from different modalities, improving processing efficiency and enhancing the ability to capture key features. The private encoder consists of invertible neural networks and transformer modules, designed to extract local and global features, respectively. A novel transformer module refines attention distribution and feature aggregation to capture superpixel-level global correlations, ensuring that the network effectively captures essential global information, thereby enhancing the quality of the fused image. Experimental results, comparing DMMFnet with nine leading fusion methods, indicate that DMMFnet significantly improves various evaluation metrics and achieves superior visual effects, demonstrating its advanced fusion capability.

Author 1: Yukun Zhang

Author 2: Lei Wang

Author 3: Muhammad Tahir

Author 4: Zizhen Huang

Author 5: Yaolong Han

Author 6: Shanliang Yang

Author 7: Shilong Liu

Author 8: Muhammad Imran Saeed

Keywords: Medical image fusion; channel-spatial attention; super token sampling; encoder–decoder

Paper 70: Deep Learning and Computer Vision-Based System for Detecting and Separating Abnormal Bags in Automatic Bagging Machines

Abstract: This paper presents a novel deep learning and computer vision-based system for detecting and separating abnormal bags within automatic bagging machines, addressing a key challenge in industrial quality control. The core of our approach is the development of a data collection system seamlessly integrated into the production line. This system captures a comprehensive variety of bag images, ensuring a dataset representative of real-world variability. To augment the quantity and quality of our training data, we implement both offline and online data augmentation techniques. For classifying normal and abnormal bags, we design a lightweight deep learning model based on the residual network for deployment on computationally constrained devices. Specifically, we improve the initial convolutional layer by utilizing ghost convolution and implement a reduced channel strategy across the network layers. Additionally, knowledge distillation is employed to refine the model's performance by transferring insights from a fully trained, more complex network. We conduct extensive comparisons with other convolutional neural network models, demonstrating that our proposed model achieves superior performance in classifying bags while maintaining high efficiency. Ablation studies further validate the contribution of each modification to the model's success. Upon deployment, the model demonstrates robust accuracy and operational efficiency in a live production environment. The system provides significant improvements in automatic bagging processes, combining accuracy with practical applicability in industrial settings.

Author 1: Trung Dung Nguyen

Author 2: Thanh Quyen Ngo

Author 3: Chi Kien Ha

Keywords: Automatic bagging machines; deep learning; computer vision; bags classification; data augmentation

Paper 71: Implementing Optimization Methods into Practice to Enhance the Performance of Solar Power Systems

Abstract: The use of contemporary technological tools, as well as the modernization of curricula in the field of electronic and electrical engineering, is one of the main objectives of the academic staff at the Universities. Efforts to improve curricula and scientific research infrastructure find strong support from the CBHE programs funded by the EU. The purpose of this work is to include the optimization methods for improving of the photovoltaic system’s performance using the digital technologies which develop students' theoretical and practical skills for a sustainable development in the field of energy. Optimizing of the photovoltaic system through the addition of a booster, MPPT controller to the existing architecture, as well as with the help of SIMULINK will increase the energy efficiency of the photovoltaic system, increasing the University's economic benefit and moreover, the ecological benefits for the population. The implementation of the optimization algorithms will increase the simulation skills of the academic staff and students for a more in-depth analysis related to the implementation of RES, an analysis which until now has only been developed through software data collection methods of the system. As case study it is the utilization of photovoltaic system in the University of Durres area, which is a sustainable development area in both the public and private sectors after the 2019 earthquake. The study brings a transdisciplinary approach that contribute in the education of the new generation towards a green society and economy. It includes knowledge of the field of electrical engineering in the direction of increasing the performance of renewable energy systems as well as the analysis of electronic circuits with the help of optimization algorithms and different ICT tools.

Author 1: Luçiana Toti

Author 2: Alma Stana

Author 3: Alma Golgota

Author 4: Eno Toti

Keywords: Optimization; photovoltaic systems; education; performance; controller; SIMULINK

Paper 72: Combined Framework for Type-2 Neutrosophic Number Multiple-Attribute Decision-Making and Applications to Quality Evaluation of Digital Agriculture Park Information System

Abstract: A digital agriculture park refers to an agricultural production and organizational unit of a certain scale where digital technology is used to optimize the agricultural supply chain. It enhances park management and service levels, achieving a new development model characterized by safe, low-carbon, high-quality, high-yield, precise, and efficient production, management, service, and operation. The quality evaluation of digital agriculture park information system is a multiple-attribute decision-making (MADM). Currently, the Exponential TODIM (ExpTODIM) and TOPSIS was put forward MADM. The Type-2 neutrosophic numbers (T2NNs) are employed to portray fuzzy information during the quality evaluation of digital agriculture park information system. In this works, the Type-2 neutrosophic number Exponential TODIM-TOPSIS (T2NN-ExpTODIM-TOPSIS) approach is put forward MAGDM under T2NNs. Finally, numerical study for quality evaluation of digital agriculture park information system is determined to demonstrate the T2NN-ExpTODIM-TOPSIS approach. The major research motivation is cultivated: (1) ExpTODIM and TOPSIS approach was enhanced under IFSs; (2) Entropy is put forward weight numbers in light with score values along with T2NNs; (3) T2NN-ExpTODIM-TOPSIS is put forward the MADM along with T2NNs; (4) numerical example for quality evaluation of digital agriculture park information system and different comparative analysis is put forward the validity of T2NN-ExpTODIM-TOPSIS.

Author 1: Wei Ji

Author 2: Ning Sun

Author 3: Botao Cao

Author 4: Xichan Mu

Keywords: Multiple-Attribute Decision-Making (MADM); Type-2 Neutrosophic Numbers (T2NNs); ExpTODIM approach; TOPSIS approach; quality evaluation

Paper 73: Using Pretrained VGG19 Model and Image Segmentation for Rice Leaf Disease Classification

Abstract: This study explores the application of the VGG19 convolutional neural network (CNN) model, pre-trained on ImageNet, for the classification of rice crop diseases using image segmentation techniques. The research aims to enhance disease detection accuracy by integrating a robust deep learning framework tailored to the specific challenges of agricultural pathology. A dataset comprising 200 images categorized into four disease classes was employed to train and validate the model. Techniques such as data augmentation, dropout, and dynamic learning rate adjustments were utilized to improve model training and prevent overfitting. The model's performance was evaluated using metrics including accuracy, precision, recall, and F1-score, with a particular focus on the ability to generalize to unseen data. Results indicated a high overall accuracy exceeding 90%, showcasing the model’s capability to effectively classify rice crop diseases. Challenges such as class-specific misclassification were addressed through the model’s learning strategy, highlighting areas for further enhancement. The research contributes to the field by demonstrating the potential of using pre-trained CNN models, adapted through fine-tuning and robust training techniques, to significantly improve disease detection in crops, thereby supporting sustainable agricultural practices and enhancing food security. Future work will explore the integration of multimodal data and real-time detection systems to broaden the applicability and effectiveness of the technology in diverse agricultural settings.

Author 1: Gulbakhram Beissenova

Author 2: Almira Madiyarova

Author 3: Akbayan Aliyeva

Author 4: Gulsara Mambetaliyeva

Author 5: Yerzhan Koshkarov

Author 6: Nagima Sarsenbiyeva

Author 7: Marzhan Chazhabayeva

Author 8: Gulnara Seidaliyeva

Keywords: Rice crop diseases; convolutional neural networks; VGG19 model; image segmentation; disease classification; data augmentation; model generalization; sustainable farming

Paper 74: Blockchain-Based Vaccination Record Tracking System

Abstract: Blockchain technology is basically a decentralized database maintained by applicable parties and has been extensively used in colorful scripts similar as logistics and finance. In terms of operations in the medical field, it's getting increasingly important because the case's symptoms may be related to a certain vaccine. Whether the case has been vaccinated with this vaccine will lead to different individual results by the croaker. This study proposes a traceable blockchain-grounded vaccination record storehouse and sharing system. In the proposed system, the case gets the vaccination at any legal clinic and the VR can be saved accompanied by the hand into the blockchain center, which ensures traceability. When the case visits the sanitarium for treatment, the croaker can gain the details of the VR from the blockchain center and also make an opinion. The security of the proposed system will be defended by the programmed smart contracts. The proper record storage after encryption will ensure data privacy, integrity and security. Blockchain tracebility uses block-chain technology to record the movement of a product in the supply chain.

Author 1: Shwetha G K

Author 2: Jayantkumar A Rathod

Author 3: Naveen G

Author 4: Mounesh Arkachari

Author 5: Pushparani M K

Keywords: Blockchain technology; decentralized; vaccine record tracking; integrity; smart contracts; vaccination record storage; traceability

Paper 75: Enhance the Security of the Cloud Using a Hybrid Optimization-Based Proxy Re-Encryption Technique Considered Blockchain

Abstract: Every day, a vast amount of data with incalculable value will be generated by the IoT devices that are deployed in various types of applications. It is crucial to ensure the reliability and safety of IoT data exchange in a cloud context because this data frequently contains the user's private information. This study presents a novel encrypted data storage and security system using the blockchain method in conjunction with hybrid optimization-based proxy re-encryption (HO-PREB). Dependency on outside central service providers is eliminated by the HO-PREB-based consensus process. In the blockchain system, several consensus nodes serve as proxy service nodes to restore encrypted data and merge transformed ciphertext with private data. Hybrid owl and bat optimization is employed to select the optimal key for enhancing security. This removes the limitations associated with securely storing and distributing private encrypted data via a distributed network. Moreover, the blockchain's distributed ledger ensures the permanent storage of data-sharing records and outcomes, ensuring accuracy and dependability. The simulated experiments of the designed model are evaluated with existing cryptographic techniques and gain a lower latency of 3.2 s and a lower turnaround time of 45 ms. Furthermore, the developed technique enhances cloud system security and possesses the capability to detect and mitigate attacks in the cloud environment.

Author 1: Ahmed I. Alutaibi

Keywords: Cloud security; Internet of Things; proxy re-encryption; blockchain; data sharing; hybrid optimization

Paper 76: Malicious Website Detection Using Random Forest and Pearson Correlation for Effective Feature Selection

Abstract: In recent years, the internet has expanded rapidly, driving significant advancements in digitalization that have transformed day to day lives. Its growing influence on consumers and the economy has increased the risk of cyberattacks. Cybercriminals exploited network misconfigurations and security vulnerabilities during these transitions. Among countless cyberattacks, phishing remains the most common form of cybercrime. Phishing via malicious Uniform Resource Locator (URL)s threatens potential victims by posing as an imposter and stealing critical and sensitive data. An increase in cyberattacks using phishing needs immediate attention to find a scalable solution. Earlier techniques like blacklisting, signature matching, and regular expression method are insufficient because of the requirement to keep updating the rule engine or signature database regularly. Significant research has recently been conducted on using Machine Learning (ML) models to detect malicious URLs. In this study, the authors have provided a study highlighting the importance of significant feature selection for training ML models for detecting malicious URLs. Pearson correlation is employed in this study for selecting significant features, and the outcome demonstrates that in terms of accuracy and other performance indices, the Random Forest classifier outperforms the other classifiers.

Author 1: Esha Sangra

Author 2: Renuka Agrawal

Author 3: Pravin Ramesh Gundalwar

Author 4: Kanhaiya Sharma

Author 5: Divyansh Bangri

Author 6: Debadrita Nandi

Keywords: Malicious URL; machine learning; feature selection; Random Forest; cybercrime

Paper 77: Enhancing Orchard Cultivation Through Drone Technology and Deep Stream Algorithms in Precision Agriculture

Abstract: The integration of cutting-edge technology in agriculture has revolutionized traditional farming practices, paving the way for Smart Agriculture. This research presents a novel approach to enhancing the cultivation of orchard crops by combining deep-stream algorithms with drone technology. Focusing on pomegranate farming, the study employs a drone system with four specialized cameras: thermal, optical RGB, multi-spectral, and LiDAR. These cameras facilitate comprehensive data collection and analysis throughout the crop growth cycle. The thermal camera monitors plant health, yield estimation, fertilizer management, and irrigation mapping. The optical RGB camera contributes to crop management by analyzing vegetation indices, assessing fruit quality, and detecting weeds. The multi-spectral and hyperspectral cameras enable early detection of crop diseases and assessment of damaged crops. LiDAR aids in understanding crop growth by measuring plant height, tracking phenology, and analyzing water flow patterns. The data collected is processed in real-time using Deep Stream algorithms on an NVIDIA Jetson GPU, providing predictive insights into key farming characteristics. Our model demonstrated superior performance compared to four established models—MDC, MLP, SVM, and ANFIS—achieving the highest accuracy (95%), sensitivity (94%), specificity (96%), and precision (91%). This integrated method offers a robust solution for precision agriculture, empowering farmers to optimize crop management, enhance productivity, and promote sustainable agriculture practices.

Author 1: P. Srinivasa Rao

Author 2: Anantha Raman G R

Author 3: Madira Siva Sankara Rao

Author 4: K. Radha

Author 5: Rabie Ahmed

Keywords: Smart agriculture; crops; cultivation; deep stream algorithms; drone and technology

Paper 78: Comparative Analysis of Small and Medium-Sized Enterprises Cybersecurity Program Assessment Model

Abstract: In the digital age, Small and Medium-sized Enterprises must review and improve their cybersecurity posture to combat rising risks. This paper thoroughly compares Small and Medium-sized Enterprises cybersecurity program assessment approaches. The National Institute of Standards and Technology's Cybersecurity Framework, CyberSecurity Readiness Model for SMEs, Cybersecurity Evaluation Model, and Adaptable Security Maturity Assessment and Standardisation framework were examined. The NIST CSF is adaptable and applicable to many sectors, while the CSRM provides a standardized way to assess an organization's cyber readiness. With its resource limits and operational scales, the CSRM-SME meets SMEs' particular issues. Organizations may examine and improve cybersecurity with CSEM. The approach can be used for SMEs, higher education institutions, and industrial control systems. The ASMAS architecture is flexible for continual security enhancement due to its scalability and standardization. This comparison analysis shows each framework's strengths and weaknesses, revealing their suitability for diverse SME scenarios. This paper helps SMEs choose the best model to strengthen cybersecurity, boost resilience, and meet global standards. This paper will compare the NIST CSF, CSRM-SME, CSEM, and ASMAS cybersecurity frameworks.

Author 1: Wan Nur Eliana Wan Mohd Ludin

Author 2: Masnizah Mohd

Author 3: Wan Fariza Paizi@Fauzi

Keywords: Cybersecurity; SMEs; cybersecurity program assessment models; cybersecurity assessment frameworks

Paper 79: Real-Time Robotic Force Control for Automation of Ultrasound Scanning

Abstract: Ablation represents a minimally invasive option for liver cancer treatment, commonly guided by imaging techniques such as ultrasound. Recently, there has been a surge in interest in semi-automated or fully automated robotic image acquisition. Specifically, there is a continuing interest in automating medical ultrasound image acquisition due to ultrasound being widely used, having a lower cost, and being more portable than other imaging modalities. This study explores automated robot-assisted ultrasound imaging for liver ablation procedures. The study proposed utilizing a collaborative robot arm from Universal Robots (UR), which has gained popularity across various medical applications. A robotic real-time force control system was designed and demonstrated to regulate the contact force exerted by the robot on the surface of a torso phantom, ensuring optimal contact during ultrasound imaging. The Robot Operating System (ROS) and the UR Real-Time Data Exchange (UR-RTDE) interface were employed to control the robot. The findings indicate that the contact force can be maintained around a set desired value of 9N. However, deviations occur due to residual forces from acceleration when the probe is not in contact with the phantom. These results provide a foundation for further advancements in the automation of ultrasound scanning.

Author 1: Ungku Muhammad Zuhairi Ungku Zakaria

Author 2: Seri Mastura Mustaza

Author 3: Mohd Hairi Mohd Zaman

Author 4: Ashrani Aizzuddin Abd Rahni

Keywords: Real-time; position control; force control; ROS; collaborative robot; admittance control; kinematics; dynamic; tool center point; damping

Paper 80: Deep Learning Model for Enhancing Automated Recycling Machine with Incentive Mechanisms

Abstract: Automated Recycling Machine (ARM) can be defined as an interactive tool to flourish recycling culture among community by providing incentive to the user that deposit the recyclable items. To enable this, the machine crucially needs a material validation module to identify the deposited recyclable items. Utilizing combination of sensors for such purpose is a tedious task and hence vision-based YOLO detection framework is proposed to identify three types of recyclable material which are aluminum can, PET bottle and tetra-pak Initially, the 14883 training samples and 937 validation samples were fed to the various YOLO variants for investigating an optimal model that can yield high accuracy and suitable for CPU usage during inference stage. Next the user interface is constructed to effectively communicate with the user when operating the ARM with easy-to-use graphical instruction. Eventually, the ARM body was designed and developed with durable material for usage in indoor and outdoor conditions. From series of experiments, it can be concluded that, the YOLOv8-m detection model well suit for the ARM material identification usage with 0.949 [email protected]:0.95 score and 0.997 F1 score. Field testing showed that the ARM effectively encourages recycling, evidenced by the significant number of recyclable items collected.

Author 1: Razali Tomari

Author 2: Aeslina Abdul Kadir

Author 3: Wan Nurshazwani Wan Zakaria

Author 4: Dipankar Das

Author 5: Muhamad Bakhtiar Azni

Keywords: Recycling machine; You Only Look Once (YOLO); vision system; interactive recycling; deep learning

Paper 81: A Hybrid of Extreme Learning Machine and Cellular Neural Network Segmentation in Mangrove Fruit Classification

Abstract: Mangroves are a collection of plants that inhabit the intertidal zone, namely the area between the lowest and highest points reached by the tide. Overall, mangroves provide a range of advantages, including the prevention of coastal erosion, the inhibition of seawater intrusion onto land leading to brackish groundwater, and serving as habitats and food sources for diverse animal species. In addition, many types of mangrove fruit have been used as sustenance for humans and as ingredients in processed food products. Mangrove fruit has a considerable variety of species, each characterized by distinct forms. At now, farmers and the general public rely only on visual observation to identify mangrove fruit species. Consequently, their ability to accurately detect the correct species is not guaranteed. In order to address this issue, this study employs digital image processing using the Extreme Learning Machine technique to facilitate the identification of various kinds and varieties of mangrove fruit by the general public and farmers. The study utilizes gray-scaling and Contrast Enhancement as image processing methods, while segmentation is performed by the use of the Cellular Neural Network approach. Following extensive testing in this study, it was determined that the used methodology effectively identified several species of mangrove fruit. The results yielded an accuracy rate of 94.11% for extracting shape, texture, and color elements, and accuracy rate of 99.63% for extracting texture and color features.

Author 1: Romi Fadillah Rahmat

Author 2: Opim Salim Sitompul

Author 3: Maya Silvi Lydia

Author 4: Fahmi

Author 5: Shifani Adriani Ch

Author 6: Pauzi Ibrahim Nainggolan

Author 7: Riza Sulaiman

Keywords: Mangroves; mangrove conservation; image processing; ecological informatics; cellular neural network; extreme learning machine

Paper 82: Evaluating the Impact of Yoga Practices to Improve Chronic Venous Insufficiency Symptoms: A Classification by Gaussian Process

Abstract: Chronic Venous Insufficiency (CVI) is a widespread condition marked by diverse venous system irregularities stemming from occlusion and varicosities. Factors like family history and lifestyle choices amplify CVI's economic consequences, emphasizing the need for proactive measures. The sedentary lifestyle of many individuals can contribute to various diseases, including CVI. Yoga is now endorsed as a multifaceted exercise to alleviate CVI symptoms, offering a holistic approach and complementary therapy for diverse medical conditions. This study developed a method for evaluating and classifying symptoms associated with varicose veins, utilizing the Venous Clinical Score (VCSS) data. A specific emphasis was placed on investigating the impact of yoga on these symptoms, and a comprehensive performance assessment was conducted based on data obtained from a cohort of 100 patients. This paper achieves optimal performance by employing the Gaussian Process Classifier (GPC) along with two optimizers, namely the Crystal Structure Algorithm (CSA) and the Fire Hawk Optimizer (FHO). The results indicate that in predicting VCSS-Pre (reflecting symptoms before engaging in yoga exercises), the GPFH exhibited superior performance with an F1-score of 0.872, surpassing the GPCS, which attained an F1-score of 0.861 by almost 1.26%. Additionally, the prediction for VCSS-1, reflecting symptoms after one month of yoga practices, revealed the GPFH outperforming the GPCS with respective F1-score values of 0.910 and 0.901.

Author 1: Feng Yun Gou

Keywords: Chronic venous insufficiency; yoga; Gaussian Process Classifier; Crystal Structure Algorithm; Fire Hawk Optimizer

Paper 83: A Semantic Segmentation Method for Road Scene Images Based on Improved DeeplabV3+ Network

Abstract: Semantic segmentation of road scenes plays a crucial role in many fields such as autonomous driving, intelligent transportation systems and urban planning. Through the precise identification and segmentation of elements such as roads, pedestrians, vehicles, and traffic signs, the system can better understand the surrounding environment and make safe and effective decisions. However, the existing semantic segmentation technology still faces many challenges in the face of complex road scenes, such as lighting changes, weather effects, different viewing angles and the existence of occlusions. Combined with the actual road scene image, this paper improves DeeplabV3+ network and applies it to semantic segmentation of road scene image, and proposes a semantic segmentation method of road scene image based on improved DeeplabV3+ network. By adding enhancement strategies for road scene images and hyperparameter adjustment, the method improves the training process of DeeplabV3+ network, and uses SK attention mechanism to improve the feature fusion module in DeeplabV3+, so as to improve the segmentation effect of road scene images. After the validation of Cityscapes and other data sets, the segmentation accuracy index mIoU of the proposed method reaches 79.8%, which can predict better semantic style effect, effectively improve the segmentation performance and accuracy of the model, and achieve better segmentation index results in the comparison network, and the subjective visual effect of the segmentation is also better.

Author 1: Lihua Bi

Author 2: Xiangfei Zhang

Author 3: Shihao Li

Author 4: Canlin Li

Keywords: Image enhancement; attention mechanism; semantic segmentation; road scene images

Paper 84: Enhancing Tuberculosis Diagnosis and Treatment Outcomes: A Stacked Loopy Decision Tree Approach Empowered by Moth Search Algorithm Optimization

Abstract: Chest X-ray imaging is the main tool for detecting tuberculosis (TB), providing essential information about pulmonary abnormalities that may indicate the presence of the disease. Still, manual interpretation is a common component of older diagnostic methods, and it may be laborious and subjective. The development of sophisticated machine learning methods offers a potential way to improve TB detection through the automation of chest X-ray image interpretation. This takes a look at goals to increase a sturdy framework for TB diagnosis the usage of Stacked Loopy Decision Trees (SLDT) optimized with the Moth Search Algorithm (MSA). The objective is to improve upon current techniques with the aid of integrating sophisticated feature extraction and ensemble mastering strategies. The novelty lies in the integration of SLDT, a hierarchical ensemble model able to shooting complex styles in chest X-ray images, with MSA for optimized parameter tuning and function selection. This technique addresses the complexity of TB analysis by enhancing each interpretability and overall performance metrics. The proposed framework employs the Gray-Level Co-prevalence Matrix (GLCM) for texture characteristic extraction, accompanied with the aid of SLDT ensemble studying optimized through MSA. This methodology objectives to discern TB-particular styles from chest X-ray pictures with excessive accuracy and efficiency. Evaluation of a comprehensive dataset demonstrates advanced performance metrics including accuracy, sensitivity, specificity, and vicinity underneath the ROC curve (AUC-ROC) compared to traditional gadget gaining knowledge of procedures. The outcomes demonstrate how well the SLDT-MSA framework performs in diagnosing TB, with 99% accuracy. The observation indicates that the SLDT-MSA framework offers practitioners a trustworthy and easily understandable solution, marking a significant advancement in TB diagnosis.

Author 1: Huma Khan

Author 2: Mithun DSouza

Author 3: K. Suresh Babu

Author 4: Janjhyam Venkata Naga Ramesh

Author 5: K. R. Praneeth

Author 6: Pinapati Lakshmana Rao

Keywords: Tuberculosis (TB); chest x-ray; stacked loopy decision trees (SLDT); moth search algorithm (MSA); medical imaging

Paper 85: Complex Environmental Localization of Scenic Spots by Integrating LANDMARC Localization System and Traditional Location Fingerprint Localization

Abstract: The scenic spot contains complex and changeable indoor and outdoor environments, some of which may be difficult to work effectively due to signal occlusion, multipath effect and other factors. In response to this problem, this paper proposes a method of Location Identification Based on the Dynamic Active Radio Frequency IDentification Calibration system and fingerprint localization system. It aims to improve positioning accuracy and reliability in the complex environment in the scenic spot. Firstly, the Location Identification Based on Dynamic Active Radio Frequency IDentification Calibration system is analyzed and improved. Then the improved positioning algorithm is applied to the complex environment of the scenic spot. Finally, the positioning results of the improved positioning algorithm in the complex environment of the scenic spot are tested. The experimental results show that when the K value is set to 4, the reader is arranged in the four corners and the center of the area, and the label density is set to 6×6, the average error of the research system in terms of error control is only 0.32, which is 0.28 less than that of the ultrasonic positioning system. All in all, the combination of Location Identification Based on Dynamic Active Radio Frequency IDentification Calibration system and traditional location fingerprint location of the scenic spot complex environment positioning scheme, it has shown great advantages in positioning accuracy, stability and real-time.

Author 1: Shasha Song

Author 2: Cong Li

Keywords: LANDMARC; localization system; fingerprint localization; environmental localization; scenic spot

Paper 86: Development of a 5G-Optimized MIMO Antenna with Enhanced Isolation Using Neutralization Line and SRRs Metamaterials

Abstract: This paper presents the design of a Multiple Input Multiple Output (MIMO) antenna intended for 5G wireless applications operating in the 3.5 GHz frequency range. The MIMO system consists of two adjacent antennas, measuring 100 mm × 80 mm × 1.6 mm, with spacing between radiating elements equal to one-eighth of the wavelength (λ/8). The antenna is constructed on an FR4 substrate with a permittivity of 4.3, and a microstrip line is employed for feeding the patch. Several techniques are employed to enhance the isolation between the antennas. Specifically, two decoupling methods are explored: the use of a neutralization line (NL) and the incorporation of metamaterial split-ring resonators (SRRs). Simulation results demonstrate substantial isolation, exceeding 20 dB with SRR implementation and more than 23 dB with the NL approach. Both individual antennas and the MIMO configuration are simulated, analyzed, and then physically fabricated for measurement, exhibiting good agreement between measured and simulated results. The study investigates the impact of each technique on antenna to determine the optimal configuration for the applications of 5G and IOT in different fields such as health (wireless medical telemetry systems (WMTS)). Remarkably, the introduction of metamaterial (MTM) with SRRs achieves a noteworthy reduction of mutual coupling by 23 dB while minimizing the mutual coupling to about 23 dB with NL insertion.

Author 1: Chaker Essid

Author 2: Linda Chouikhi

Author 3: Alsharef Mohammad

Author 4: Bassem Ben Salah

Author 5: Hedi Sakli

Keywords: 5G; antenna; MIMO; SRRs metamaterials; isolation; IOT; neutralization line (NL)

Paper 87: Deep Learning Fusion for Intracranial Hemorrhage Classification in Brain CT Imaging

Abstract: Brain hemorrhages are characterized by the rupture in the arteries of brain due blood clotting or high blood pressure (BP), presents a significant risk of traumatic injury or even death. This bleeding results in the damage in brain cells, with common causes including brain tumors, aneurysm, blood vessel abnormalities, amyloid angiopathy, trauma, high BP, and bleeding disorders. When a hemorrhage happens, oxygen can no longer reach the brain tissues and brain cells begin to die if they are depleted of oxygen and nutrients for longer than three or four minutes. The affected nerve cells and the related functions they control are damaged as well. Early detection of brain hemorrhages is crucial. In this paper an efficient hybrid deep learning (DL) model is proposed for the intracranial hemorrhage detection (ICH) from brain CT images. The proposed method integrates DenseNet 121 and Long Short-Term Memory (LSTM) models for the accurate classification of ICH. The DenseNet 121 model act as the feature extraction model. The experimental results demonstrated that the model attained 97.50% accuracy, 97.00% precision, 95.99% recall and 96.33% F1 score, demonstrating its effectiveness in accurately identifying and classifying ICH.

Author 1: Padma Priya S. Babu

Author 2: T. Brindha

Keywords: Intracranial hemorrhage; deep learning; DenseNet 121; LSTM; brain CT images

Paper 88: Romanian Sign Language and Mime-Gesture Recognition

Abstract: This paper presents a comprehensive approach to Romanian Sign Language (RSL) recognition using machine learning techniques. The primary focus is on developing and evaluating a robust model capable of accurately classifying hand and mime gestures representative of RSL and converting it into speech through an application. Utilizing a dataset of hand landmarks captured and stored in CSV format, the study outlines the preprocessing steps, model training, and performance evaluation. Key components of the methodology include data preparation, model training, performance evaluation and model optimization. The results demonstrate the feasibility of using machine learning for RSL recognition, achieving promising accuracy rates. The study concludes with a discussion on potential applications and future enhancements, including real-time gesture recognition and expanding the dataset for improved generalization. This work contributes to the broader effort of making sign language more accessible through technology, particularly for the Romanian-speaking deaf and hard-of-hearing community.

Author 1: Enachi Andrei

Author 2: Turcu Cornel

Author 3: George Culea

Author 4: Sghera Bogdan Constantin

Author 5: Ungureanu Andrei Gabriel

Keywords: RSL; sign language; machine learning; model; mime gestures

Paper 89: Multiclass Osteoporosis Detection: Enhancing Accuracy with Woodpecker-Optimized CNN-XGBoost

Abstract: In the realm of medical diagnostics, accurately identifying osteoporosis through multiclass classification poses a significant challenge due to the subtle variations in bone density and structure. This study proposes a novel approach to enhance detection accuracy by integrating the Woodpecker Optimization Algorithm with a hybrid Convolutional Neural Network (CNN) and XGBoost model. The Woodpecker Optimization Algorithm is employed to fine-tune the CNN-XGBoost model parameters, leveraging its ability to efficiently search for optimal configurations amidst complex data landscapes. The proposed framework begins with the CNN component, designed to automatically extract hierarchical features from bone density images. This initial stage is crucial for capturing intricate patterns that signify osteoporotic conditions across multiple classes. Subsequently, the extracted features are fed into an XGBoost classifier, renowned for its robust performance in handling structured data and multiclass classification tasks. By combining these two powerful techniques, the model aims to synergistically utilize the strengths of deep learning in feature extraction and gradient boosting in decision-making. Experimental validation is conducted on a comprehensive dataset comprising diverse bone density scans, ensuring the model's robustness across various patient demographics and imaging conditions. Performance criteria including recall, precision, reliability, and F1-score are assessed to show how well the suggested Woodpecker-optimized CNN-XGBoost framework performs in comparison to other approaches when it comes to obtaining better accuracy in diagnosis. The findings underscore the potential of hybrid models in advancing osteoporosis detection, offering clinicians a reliable tool for early and precise diagnosis, thereby facilitating timely interventions to mitigate the debilitating effects of bone-related diseases. Osteoporosis detection model with a classification accuracy of 97.1% implemented in Python.

Author 1: Mithun DSouza

Author 2: Divya Nimma

Author 3: Kiran Sree Pokkuluri

Author 4: Janjhyam Venkata Naga Ramesh

Author 5: Suresh Babu Kondaveeti

Author 6: Lavanya Kongala

Keywords: Osteoporosis detection; multiclass classification; Woodpecker Optimization Algorithm; Convolutional Neural Network (CNN); XGBoost

Paper 90: Attention-Based Joint Learning for Intent Detection and Slot Filling Using Bidirectional Long Short-Term Memory and Convolutional Neural Networks

Abstract: Effective natural language understanding is crucial for dialogue systems, requiring precise intent detection and slot filling to facilitate interactions. Traditionally, these subtasks have been addressed separately, but their interconnection suggests that joint solutions yield better results. Recent neural network-based approaches have shown significant performance in joint intent detection and slot filling tasks. The two primary neural network structures used are recurrent neural networks (RNNs) and convolutional neural networks (CNNs). RNNs capture long-term dependencies and store previous information semantics in a fixed-size vector, but their ability to extract global semantics is limited. CNNs can capture n-gram features using convolutional filters, but their performance is constrained by filter width. To leverage the strengths and mitigate the weaknesses of both networks, this paper proposes an attention-based joint learning classification for intent detection and slot filling using BiLSTM and CNNs (AJLISBC). The BiLSTM encodes input sequences in both forward and backward directions, producing high-dimensional representations. It applies scalar and vectorial attention to obtain multichannel representations, with scalar attention calculating word-level importance and vectorial attention assessing feature-level importance. For classification, AJLISBC employs a CNN structure to capture word relations in the representations generated by the attention mechanism, effectively extracting n-gram features. Experimental results on the benchmark Airline Travel Information System (ATIS) dataset demonstrate that AJLISBC outperforms state-of-the-art methods.

Author 1: Yusuf Idris Muhammad

Author 2: Naomie Salim

Author 3: Sharin Hazlin Huspi

Author 4: Anazida Zainal

Keywords: Joint learning; intent detection; slot filling; multichannel

Paper 91: Attention-Based Deep Learning Approach for Pedestrian Detection in Self-Driving Cars

Abstract: Autonomous vehicle safety relies heavily on the ability to accurately detect pedestrians, as this capability is crucial for preventing accidents and saving lives. Pedestrian recognition is particularly challenging in the dynamic and complex environments of urban areas. Effective pedestrian detection is crucial for ensuring road safety in autonomous vehicles. Current pedestrian identification systems often fall short in capturing the nuances of pedestrian behavior and appearance, potentially leading to dangerous situations. These limitations are mainly due to difficulties in various conditions, such as low-light environments, occlusions, and intricate urban settings. This paper proposes a novel solution to these challenges by integrating an attention-based convolutional bi-GRU model with deep learning techniques for pedestrian recognition. This method leverages deep learning to provide a robust solution for pedestrian detection. Convolutional layers are utilized to extract spatial features, attention mechanisms highlight semantic details, and Bidirectional Gated Recurrent Units (Bi-GRU) capture the temporal context in the proposed model. The process begins with data collection to build a comprehensive pedestrian dataset, followed by preprocessing using min-max normalization. The key components of the model work together to enhance pedestrian detection, ensuring a more accurate and comprehensive understanding of dynamic pedestrian scenarios. The implementation of this unique approach was carried out using Python, employing libraries such as TensorFlow, Keras, and OpenCV. The proposed attention-based convolutional bi-GRU model outperforms previous models by an average of 17.1%, achieving an accuracy rate of 99.4%. The model significantly surpasses Random Forest, Faster R-CNN, and SVM in terms of pedestrian recognition accuracy, which is critical for autonomous vehicle safety.

Author 1: Wael Ahmad AlZoubi

Author 2: Girish Bhagwant Desale

Author 3: Sweety Bakyarani E

Author 4: Uma Kumari C R

Author 5: Divya Nimma

Author 6: K Swetha

Author 7: B Kiran Bala

Keywords: Pedestrian recognition; autonomous vehicle safety; deep learning; attention mechanism; Bidirectional Gated Recurrent Units

Paper 92: Cryptographic Techniques in Digital Media Security: Current Practices and Future Directions

Abstract: Content privacy and unauthorized access to copyrighted digital media content are common in the dynamic, fast-paced digitalized media marketplace. Cryptographic methods are the foundation of modern digital media security, and they must ensure the security, integrity, and authenticity of digital media data. This article analyses cryptographic methods that are used to protect digital media content. The paper reviews the main cryptographic concepts, such as symmetric cryptography, asymmetric cryptography, hash functions, and digital signatures. The paper also discusses some popular approaches: encryption, Digital Rights Management (DRM), watermarking, and solutions based on blockchain. Finally, we highlight implementation challenges such as key management and scalability and identify emerging trends such as quantum-safe cryptography and privacy-preserving techniques. By presenting the current research results and discussing the directions for the future, the study aims to pave the way for secure, efficient, and robust cryptographic solutions for digital media protection, leading to sustainable development, innovation, and secure communication of digital content among users.

Author 1: Gongling ZHANG

Keywords: Digital media; cryptographic; content security; digital rights management; watermarking; blockchain

Paper 93: Detecting Online Gambling Promotions on Indonesian Twitter Using Text Mining Algorithm

Abstract: This study addresses the pressing challenge of detecting online gambling promotions on Indonesian Twitter using text mining algorithms for text classification and analytics. Amid limited research on this subject, especially in the Indonesian context, we aim to identify common textual features used in gambling promotions and determine the most effective classification models. By analyzing a dataset of 6038 tweets collected and using methods such as Random Forest, Logistic Regression, and Convolutional Neural Networks, complemented by a comparison analysis of text representation methods, we identified frequently occurring words such as 'link', 'situs', 'prediksi', 'jackpot', 'maxwin', and 'togel'. The results indicate that the combination of TF-IDF and Random Forest is the most effective method for detecting online gambling promotion content on Indonesian Twitter, achieving a recall value of 0.958 and a precision value of 0.966. These findings can contribute to cybersecurity and support law enforcement in mitigating the negative effects of such promotions, particularly on the Twitter platform in Indonesia.

Author 1: Reza Bayu Perdana

Author 2: Ardin

Author 3: Indra Budi

Author 4: Aris Budi Santoso

Author 5: Amanah Ramadiah

Author 6: Prabu Kresna Putra

Keywords: Social media; analytics; online gambling; intention classification

Paper 94: Securing RPL Networks with Enhanced Routing Efficiency with Congestion Prediction and Load Balancing Strategy

Abstract: Low power and Lossy Networks (LLNs) are essential components of the Internet of Things (IoT) environment. In LNNs, the Routing Protocol for LLN (RPL)-based Internet Protocol Version 6 (IPv6) routing protocol is regarded as a standardized solution. However, the existing models did not account for the issues with congestion and security when modeling the RPL. Thus, to resolve these issues, this paper proposes a novel Exponential Poisson Distribution–Fuzzy (EPD-Fuzzy) model and Kullback Leibler Divergence-based Tunicate Swarm Algorithm (KLD-TSA) for developing a reliable RPL model. The hash codes are first generated for the registered nodes at the network end in order to achieve security; the hash codes are subsequently compared via requests with the immediate nodes. Each node sends a request to its neighbors using the hash value; if the hash value matches, a path is formed. The parent nodes are then chosen and ranked using the Pearson Correlation Coefficient-Spotted Hyena Optimization Algorithm (PCC-SHOA) technique to minimize latency. To avoid congestion, the EPD-Fuzzy is employed to predict congestion; then, a genitor node is introduced in the congested scenarios. The big data and videos are split, compressed, and sent via multiple paths to reduce the losses in the RPL. Moreover, to avoid network traffic, a novel KLD-TSA load balancing is introduced at the user end. The experiential outcomes exhibited the proposed technique’s effectiveness regarding Packet delivery ratio (PDR).

Author 1: Saumya Raj

Author 2: Rajesh R

Keywords: Low power and Lossy Network (LLN); Routing Protocol for LLN (RPL); load balancing; Internet of Things (IoT); Internet Protocol Version 6 (IPv6)

Paper 95: Optimizing Hyperparameters in Machine Learning Models for Accurate Fitness Activity Classification in School-Aged Children

Abstract: Classification using machine learning algorithms in physical fitness tests carried out by students in educational centers can help prevent obesity and other related diseases. This research aims to evaluate physical fitness using percentiles of the tests and machine learning algorithms with hyperparameter optimization. The process followed was knowledge discovery in databases (KDD). Data were collected from 1525 students (784 women, 741 men) aged 6 to 17, selected non-probabilistically from five public schools. For the evaluation, anthropometric parameters such as age, weight, height, sitting height, abdominal circumference, relaxed arm circumference, oxygen saturation, resting heart rate, and maximum expiratory flow were considered. Physical Fitness tests included sitting flexibility, kangaroo horizontal jump, and 20-meter fly speed. Within the percentiles observed, we took three cut-off points as a basis for the present research: > P75 (above average), p25 to p75 (average), and < P25 (below average). The following machine learning algorithms were used for classification: Random Forest, Support Vector Machine, Decision tree, Logistic Regression, Naive Bayes, K-nearest neighbor, XGBboost, Neural network, Cat Boost, LGBM, and Gradient Boosting. The algorithms were hyperparameter optimized using GridSearchCV to find the best configurations. In conclusion, the importance of hyperparameter optimization in improving the accuracy of machine learning models is highlighted. Random Forest performs well in classifying the “High” and “Low” categories in most tests but struggles to correctly classify the “Normal” category for both male and female students.

Author 1: Britsel Calluchi Arocutipa

Author 2: Magaly Villegas Cahuana

Author 3: Vanessa Huanca Hilachoque

Author 4: Marco Cossio Bolaños

Keywords: Machine learning; classification; physical fitness; schoolchildren; hyperparameters

Paper 96: Modeling Micro Traffic Flow Phenomena Based on Vehicle Types and Driver Characteristics Using Cellular Automata and Monte Carlo

Abstract: The modeling of micro traffic flow on a highway has been extensively observed and studied in various aspects, such as driver characteristics in car-following and lane-changing behaviors. Regarding car-following and lane-changing, an interesting aspect is how to model the movement conditions of vehicles on a highway that exhibit unique characteristics regarding the speed of four-wheeled or more vehicles passing through it. This condition occurs on the Porong Highway in Sidoarjo, East Java, Indonesia. Based on these conditions, this study develops a microscopic traffic flow model incorporating driver characteristics categorized into three types: careful drivers, ordinary drivers, and skilled drivers, each with distinct vehicle speed traits. These driver characteristics are integrated into the Nagel-Schreckenberg Stochastic Traffic Cellular Automata (NaSch STCA) model, which we refer to as the Modified NaSch STCA. The Monte Carlo simulation is employed to generate events through random numbers for the Occupied Initial State, Slowdown Probability, and Probability of Lane Changing. These three components are integral parts of the Modified NaSch STCA model. Experiments (simulations) were conducted on the constructed vehicle movement model, and one of the outcomes is that the travel time obtained from the NaSch STCA model is significantly faster than that obtained from the Modified NaSch STCA model. This condition is attributed to the unique vehicle speed characteristics on the Porong Highway, where the average speed vr = 38 km/h is relatively lower than the average speed typically observed on a highway.

Author 1: Tri Harsono

Author 2: Kohei Arai

Keywords: Micro traffic flow; driver characteristics; cellular automata; Monte Carlo

Paper 97: Convolutional Neural Network Model for Cacao Phytophthora Palmivora Disease Recognition

Abstract: Cacao, scientifically known as Theobroma cacao, is a highly nutritious food and is extensively utilized in multiple sectors, including agriculture and health. Nevertheless, the agricultural sector encounters notable obstacles as a result of Cacao disease such as pod rot, predominantly attributed to the Phytophthora genus. The objective of this work is to conduct a comparative analysis to determine the most effective machine-learning technique for the detection of P. palmivora infection in Cacao pods. Few studies have delved into this topic previously, but this study focuses in utilizing a little larger dataset, achieving better model, and attaining higher accuracy. A total of 2000 images of cacao pods, both healthy and disease-infected were collected. Subsequently, the images were subjected to manual classification by a domain expert based on the discernible presence or absence of the disease. The study examined six machine learning algorithms, specifically Naïve Bayes, Random Forest, Hoeffding Tree, Multilayer Neural Network, and Convolutional Neural Network (CNN). The CNN model had 99% level of accuracy, the highest among the five machine learning algorithms in the testing phase. The methodology has the potential to significantly advance sustainable agricultural practices and disease management. To enhance the model's recognition capabilities, additional datasets encompassing a broader range of Cacao varieties is necessary.

Author 1: Jude B. Rola

Author 2: Jomari Joseph A. Barrera

Author 3: Maricel V. Calhoun

Author 4: Jonah Flor Oraño – Maaghop

Author 5: Magdalene C. Unajan

Author 6: Joshua Mhel Boncalon

Author 7: Elizabeth T. Sebios

Author 8: Joy S. Espinosa

Keywords: Machine-learning; Convolutional Neural Network; detection of P. palmivora

Paper 98: Leveraging Mechanomyography Signal for Quantitative Muscle Spasticity Assessment of Upper Limb in Neurological Disorders Using Machine Learning

Abstract: Upper motor neuron syndrome is characterised by spasticity, which represents a neurological disability that can be found in several disorders such as cerebral palsy, amyotrophic lateral sclerosis, stroke, brain injury, and spinal cord injury. Muscle spasticity is always assessed by therapists using conventional methods involving passive movement and assigning spasticity grades to the relevant joints based on the degree of muscle resistance which leads to inconsistency in assessment and could affect the efficiency of the rehabilitation process. To address this problem, the study proposed to develop a muscle spasticity model using Mechanomyography (MMG) signals from the forearm muscles. The muscle spasticity model leveraged based on the Modified Ashworth Scale and focus on flexion and extension movements of the forearm. Thirty subjects who satisfied the requirements and provided consent were recruited to participate in the data collection. The data underwent a pre-processing stage and was subsequently analysed prior to the extraction of features. The dataset consists of forty-eight extracted features from the three-direction x, y, z axes (for both biceps and triceps muscle), corresponding to the longitudinal, lateral, and transverse orientations relative to the muscle fibers. Significant features selection was conducted to analyse if overall significant difference showed in the combined set of these features across the different spasticity levels. The test results determined the selection of twenty-five features from a total of forty-eight which be used to train an optimum classifier algorithm for the purpose of quantifying the level of muscle spasticity. Linear Discriminant Analysis (LDA), Decision Trees (DTs), Support Vector Machine (SVM), and K-Nearest Neighbour (KNN) algorithms have been employed to achieve better accuracy in quantifying the muscle spasticity level. The KNN-based classifier achieved the highest performance, with an accuracy of 91.29% with k=15, surpassing the accuracy of other classifiers. This leads to consistency in spasticity evaluation, hence offering optimum rehabilitation strategies.

Author 1: Muhamad Aliff Imran Daud

Author 2: Asmarani Ahmad Puzi

Author 3: Shahrul Na’im Sidek

Author 4: Ahmad Anwar Zainuddin

Author 5: Ismail Mohd Khairuddin

Author 6: Mohd Azri Abdul Mutalib

Keywords: Spasticity; mechanomyography; Modified Ashworth Scale; machine learning

Paper 99: Interactive Color Design Based on AR Virtual Implantation Technology Between Users and Artificial Intelligence

Abstract: To achieve user interactive color design, this study takes furniture color interactive design as an example, introduces artificial intelligence and augmented reality virtual implantation technology, allowing users to design furniture colors and styles according to their own ideas. By improving cyclic consistency to generate adversarial networks, furniture image style transfer is carried out, and indoor feature point classification and virtual model registration are carried out through density based spatial clustering and other methods to design an unlabeled augmented reality furniture system. The results showed that compared to other methods, the improved cyclic consistency generation adversarial network had a higher structural similarity value. In the zebra to horse image, the structural similarity value was 0.987, which was 0.018 higher than the algorithm before improvement. The registration effect of density-based spatial clustering algorithm was good, with a shorter time consumption in different scenarios, and a maximum time consumption of 0.308 seconds in occluded composite scenes. The performance of the drawing component is good, with each process of tracking threads taking less than 20ms. The research method not only satisfies users in designing furniture colors and styles, but also enhances their experience.

Author 1: Jun Ma

Author 2: Ying Chen

Keywords: Artificial intelligence; AR virtual implantation technology; color; style transfer; furniture

Paper 100: A Comprehensive Authentication Taxonomy and Lightweight Considerations in the Internet-of-Medical-Things (IoMT)

Abstract: The potential of Internet-of-Things (IoT) in healthcare is evident in its ability to connect medical equipment, sensors, and healthcare personnel to provide high-quality medical expertise in remote locations. The constraints faced by these devices such as limited storage, power, and energy resources necessitate the need for a lightweight authentication mechanism that is both efficient and secure. This study contributes by exploring challenges and lightweight authentication advancement, focusing on their efficiency on the Internet-of-Medical-Things (IoMT). A review of recent literature reveals ongoing issues such as the high complexity of cryptographic operations, scalability challenges, and security vulnerabilities in the proposed authentication systems. These findings lead to the need for multi-factor authentication with a simplified cryptographic process and more efficient aggregated management practices tailored to the constraints of IoMT environments. This study also introduces an extended taxonomy, namely, Lightweight Aggregated Authentication Solutions (LAAS), a lightweight efficiency approach that includes a streamlined authentication process and aggregated authentication, providing an understanding of lightweight authentication approaches. By identifying critical research gaps and future research directions, this study aims to provide a secure authentication protocol for IoMT and similar resource-constraint domains.

Author 1: Azlina binti Ahmadi Julaihi

Author 2: Md Asri Ngadi

Author 3: Raja Zahilah binti Raja Mohd Radzi

Keywords: Lightweight authentication; Aggregated Authentication’ Multi-Factor Authentication (MFA); Internet-of-Medical Things (IoMT)

Paper 101: Dynamic Simulation and Forecasting of Spatial Expansion in Small and Medium-Sized Cities Using ANN-CA-Markov Models

Abstract: This study utilizes the ANN-CA-Markov (Artificial Neural Network-Cellular Automata-Markov) model to address spatial planning and expansion challenges in China’s small and medium-sized cities. With China’s urbanization rate reaching 59.58% in 2018 and expected to hit 70% by 2030, the country is entering a mid-stage of urbanization, leading to rapid expansion of megacities and a gradual decline in smaller cities. The study aims to dynamically simulate urban spatiotemporal evolution and predict future land use changes, integrating land use data, DEM elevation, transportation, administrative centers, and ecological information. The model forecasts the ecological spatial layout of Wanzhou District by 2025, with results indicating a slight decrease in ecological space and an increase in construction land. This suggests a need to balance urban development with ecological sustainability amidst rapid urbanization. The study demonstrates the high accuracy of the ANN-CA-Markov model in predicting land use changes and provides valuable insights for urban planners in making informed land use decisions.

Author 1: Chengquan Gao

Keywords: ANN-CA; Markov; small and medium-sized cities; spatial; planning

Paper 102: Advanced IoT-Enabled Indoor Thermal Comfort Prediction Using SVM and Random Forest Models

Abstract: Predicting thermal comfort within indoor environments is essential for enhancing human health, productivity, and well-being. This study uses interdisciplinary approaches, integrating insights from engineering, psychology, and data science to develop sophisticated machine learning models that predict thermal comfort. Traditional methods often depend on subjective human input and can be inefficient. In contrast, this research applies Support Vector Machines (SVM) and Random Forest algorithms, celebrated for their precision and speed in handling complex datasets. The advent of the Internet of Things (IoT) further revolutionizes building management systems by introducing adaptive control algorithms and enabling smarter, IoT-driven architectures. We focus on the comparative analysis of SVM and Random Forest in predicting indoor thermal comfort, discussing their respective advantages and limitations under various environmental conditions and building designs. The dataset we used included comprehensive thermal comfort data, which underwent rigorous preprocessing to enhance model training and testing—80% of the data was used for training and the remaining 20% for testing. The models were evaluated based on their ability to accurately mirror complex interactions between environmental factors and occupant comfort levels. The results indicated that while both models performed robustly, Random Forest demonstrated greater stability and slightly higher accuracy in most scenarios. The paper proposes potential strategies for incorporating additional predictive features to further refine the accuracy of these models, emphasizing the promise of machine learning in advancing indoor comfort optimization.

Author 1: Nurtileu Assymkhan

Author 2: Amandyk Kartbayev

Keywords: Heating; building energy management; thermal comfort; IoT; Support Vector Machine; Random Forest

Paper 103: Exploration of Deep Semantic Analysis and Application of Video Images in Visual Communication Design Based on Multimodal Feature Fusion Algorithm

Abstract: Fully utilizing image and video semantic processing techniques can play a more effective role in visual communication design. In order to further explore the application of multimodal feature fusion algorithm (MFF) in video image feature analysis in visual communication design, with the aim of enhancing the depth and breadth of design creation. This article focuses on the application of video semantic understanding technology by combining image and video semantic processing techniques, in order to achieve a comprehensive, three-dimensional, and open expansion of design thinking. The MFF algorithm was proposed and implemented, which innovatively integrates multimodal information such as visual and audio in videos, deeply explores action semantics, and shows significant performance improvements compared to traditional algorithms. Specifically, compared to the other two mainstream algorithms, its performance has improved by 24.33% and 14.58%, respectively. This discovery not only validates the superiority of MFF algorithm in the field of video semantic understanding, but also reveals the profound impact of video semantic understanding technology on visual communication design practice, providing new perspectives and tools for the design industry and promoting innovation and development of design thinking. The novelty of this study lies in its interdisciplinary methodology, which applies advanced algorithm techniques to the field of art and design, and the significant improvement of the proposed MFF algorithm in enhancing design efficiency and creativity.

Author 1: Yanlin Chen

Author 2: Xiwen Chen

Keywords: Video semantics; understanding; visual communication; design

Paper 104: A Deep Reinforcement Learning (DRL) Based Approach to SFC Request Scheduling in Computer Networks

Abstract: This study investigates the use of Deep Reinforcement Learning (DRL) to minimize the latency between the source and destination of Service Function Chaining (SFC) requests in Neural Networks. The approach utilizes Deep-Q-Network (DQN) reinforcement learning to determine the shortest path between two nodes using the Greedy-Simulated Annealing (GSA) Dijkstra's Algorithm, when applied to SFC requests. The containers within the SFC framework help train the RL model based on bandwidth restrictions (fiber networks) to optimize the different pathways in terms of action space. Through rigorous evaluation of varying action spaces in models, we assessed that the Dijikstra’s Algorithm, within the sphere, is in fact a viable optimized solution to SFC request based problems. Our findings illustrate how this framework can be applied to early request based topologies to introduce a more optimized method of resource allocation, quality of service, and network performance to generalize the relationship between SFC and RL.

Author 1: Eesha Nagireddy

Keywords: RL models; SFC chain; Deep-Q-Network; Dijkstra’s algorithm

Paper 105: Improving Automatic Short Answer Scoring Task Through a Hybrid Deep Learning Framework

Abstract: An automatic short-answer scoring system involves using computational techniques to automatically evaluate and score student answers based on a given question and desired answer. The increasing reliance on automated systems for assessing student responses has highlighted the need for accurate and reliable short-answer scoring mechanisms. This research aims to improve the understanding and evaluation of student answers by developing an advanced automatic scoring system. While previous studies have explored various methodologies, many fail to capture the full complexity of response text. To address this gap, our study combines the strengths of classical neural networks with the capabilities of large language models. Specifically, we fine-tune the Bidirectional Encoder Representations from Transformers (BERT) model and integrate it with a recurrent neural network to enhance the depth of text comprehension. We evaluate our approach on the widely-used Mohler dataset and benchmark its performance against several baseline models using RMSE (Root Mean Square Error) and Pearson correlation metrics. The experimental results demonstrate that our method outperforms most existing systems, providing a more robust solution for automatic short-answer scoring.

Author 1: Soumia Ikiss

Author 2: Najima Daoudi

Author 3: Manar Abourezq

Author 4: Mostafa Bellafkih

Keywords: Student answer; automatic scoring; BERT language model; LSTM neural network; Natural Language Processing

Paper 106: BlockChain and Deep Learning with Dynamic Pattern Features for Lung Cancer Diagnosis

Abstract: Cancers in the respiratory tract grow out of control in lung carcinoma, a deadly disease. Because cancers have irregular shapes, it can be challenging to diagnose them and determine their sizes and forms from imaging studies. Furthermore, a serious issue with health image inquiry is large disparity. Artificial intelligence and blockchain are two cutting-edge advances in the healthcare industry. This paper introduces a Blockchain with a deep learning network for the early diagnosis of lung cancer in an effort to address these problems. Images from CT scans and CXRs were included in the LIDC-IDRI and NIH Chest X-ray collection. Initially, these images are pre-processed by Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance the image clarity and reduce the noise. Then the Honey Badger optimization Algorithm (HBA) is used to segment the lung region from the pre-processed image. Morphological segments of the lung region are used to generate dynamic patterns. Finally, these patterns are aggregated into the deep neural Spiking Convolutional Neural Network (SCNN), which is the global model for classifying the images into normal and abnormal cases. Based on the classification, the SCNN model achieves 98.64% accuracy from the LIDC-IDRI database and 98.9% on the NH Chest X-ray lung image dataset. The experiments indicate that the proposed approach results in lower energy consumption and faster inference times. Furthermore, the interpretability of the classification findings is improved by the intrinsic explainability of SCNNs, offering more profound understanding of the decision-making process. With these benefits, SCNNs are positioned as a reliable and effective technique for classifying lung images, providing a significant advancement over current methods.

Author 1: A. Angel Mary

Author 2: K. K. Thanammal

Keywords: Lung cancer; spiking convolutional neural network; LIDC-IDRI; CLAHE; Honey Badger optimization Algorithm; segmentation; classification

Paper 107: Application of Improved CSA Algorithm-Based Fuzzy Logic in Computer Network Control Systems

Abstract: In the past few years, with the high-speed popularization of computers and the widespread use of smart phones and mobile devices, the Internet has gradually become an indispensable part of people's daily lives. The Internet is constantly driving the process of digital society and providing people with more convenient and innovative applications. However, the internet industry also faces challenges such as runtime ambiguity, instability, large data volume, and difficulties in network situational awareness. In response to the above issues, this study combines the standard cuckoo algorithm with a fuzzy neural network to design a computer network situational awareness system. It uses principal component analysis to deduct the dimensions of the original data and then adds Gaussian noise to introduce appropriate randomness. The test proved that the improved model had a significant optimization effect on real network data, with an improvement of about 81.2% compared to the standard cuckoo algorithm. In the 220th iteration of the test set, the Loss function value was 0, which could accurately predict the network situation, with an accuracy rate of 98%. The designed system identification has higher recognition accuracy and less time consumption and has certain application potential in computer networks.

Author 1: Jianxi Yu

Keywords: CSA; computer network; fuzzy logic; principal component analysis method; network operation; situation awareness

Paper 108: Application of Sanda-Assisted Teaching System Integrating VR Technology from a 5G Perspective

Abstract: Since the focus of Sanda teaching is to allow students to master martial arts techniques through confrontational exercises, Sanda teaching in colleges and universities commonly adopts contextualized teaching methods. However, this Sanda teaching method suffers from deficiencies such as poor teaching effectiveness and difficulty in reflecting the effectiveness of Sanda martial arts. In order to solve these problems and make up for the shortcomings of offline Sanda teaching, the study adopts the virtual reality technology and the fifth generation mobile communication technology to construct a Sanda-assisted teaching system for college students. In order to ascertain whether students complete the Sanda movement practice, the study has designed two models: a self-supervised model based on acceleration and angular velocity contrast learning and a multi-task semi-supervised model based on time-frequency contrast learning. These models aim to improve the analytical function of the Sanda-assisted teaching system and address the analytical deficiencies of the existing human movement identification algorithms. The results indicated that the maximum accuracy of the research-designed self-supervised model was 95.76% and 95.89% on the training and test sets, respectively. The multi-task semi-supervised model designed in the study plateaued after nearly 22 and 24 iterations on the training and test sets, respectively. The average response time of the research-designed system was 59ms, and the throughput could reach a maximum of 77651bit/s. The model and the research-designed system both worked well, and they can lower the risk of student injuries while offering technological support for Sanda-assisted teaching and learning in higher education institutions.

Author 1: Zhaoquan Zhang

Author 2: Yong Ding

Keywords: VR technology; Sanda; teaching system; motion recognition; feature extraction

Paper 109: Data Collection Method Based on Data Perception and Positioning Technology in the Context of Artificial Intelligence and the Internet of Things

Abstract: Wireless sensor networks are an important technical form of the underlying network of the Internet of Things. The energy of each node in the network is finite. When a node runs out of energy, it can cause network interruptions, which can affect the reliability of data collection. To reduce the consumption of communication resources and ensure the reliability of data collection, the study proposes data collection based on data compression perception positioning technology. This method first uses a Bayesian compression perception method to select nodes, and then adopts an adaptive sparse strategy to collect data. When selecting nodes using this proposed method, wireless sensor networks had the longest network lifespan. In the case of different degrees of redundancy and sparsity, the research method had the lowest reconstruction error, with reconstruction errors of 0.31 and 0.40, respectively. When the balance factor was set to 0.6, the reconstruction error of the research method was the lowest, with a minimum reconstruction error of 0.05. This proposed method has better reconstruction performance, effectively prolongs the lifespan of wireless sensor networks, and reduces the consumption of communication resources.

Author 1: Xinbo Zhao

Author 2: Fei Fei

Keywords: Wireless sensor network; data collection; compression perception technology; Sparse Bayesian Learning; signal reconstruction

Paper 110: Hyperparameter Optimization in Transfer Learning for Improved Pathogen and Abiotic Plant Disease Classification

Abstract: The application of machine learning, particularly through image-based analysis using computer vision techniques, has greatly improved the management of crop diseases in agriculture. This study explores the use of transfer learning to classify both spreadable and non-spreadable diseases affecting soybean, lettuce, and banana plants, with a special focus on various parts of the banana plant. In this research, 11 different transfer learning models were evaluated in Keras, with hyperparameters such as optimizers fine-tuned and models retrained to boost disease classification accuracy. Results showed enhanced detection capabilities, especially in models like VGG_19 and Xception, when optimized. The study also proposes a new approach by integrating an EfficientNetV2-style architecture with a custom-designed activation function and optimizer to improve model efficiency and accuracy. The custom activation function combines the advantages of ReLU and Tanh to optimize learning, while the hybrid optimizer merges feature of Adam and Stochastic Gradient Descent (SGD) to balance adaptive learning rates and generalization. This innovative approach achieved outstanding results, with an accuracy of 99.96% and an F1 score of 0.99 in distinguishing spreadable and non-spreadable plant diseases. The combination of these advanced methods marks a significant step forward in the use of machine learning for agricultural challenges, demonstrating the potential of customized neural network architectures and optimization strategies for accurate plant disease classification.

Author 1: Asha Rani K P

Author 2: Gowrishankar S

Keywords: Spreadable diseases; non-spreadable diseases; transfer learning; Keras; optimizers; CNN; underfitting and overfitting; retraining the models; base models; finetuning; abiotic; biotic; infectious and non-infectious diseases; custom optimization techniques; hyperparameter tuning in neural networks; hybrid activation functions

Paper 111: Design and Application of Intelligent Visual Communication System for User Experience

Abstract: The design and application of visual communication system should be human-oriented, but currently this is often ignored by designers, resulting in poor user experience of visual communication. In order to improve the experience effect of visual communication system, combined with the existing computer technology, this paper proposes an intelligent visual communication system for user experience. First, for the problem of extracting multimodal features of users, considering the characteristics of different modal data, long and short-term memory networks are used to extract features with contextual information, and multi-scale convolutional neural networks are used for visual modality to extract low-level features from video frames. In the cross-modal stage, the low-level features in the source modality are used to enhance the target modality features. Then, for the personalized recommendation problem of users, a graph information extractor is constructed based on the graph convolutional neural network to fuse the recommended user-item bipartite graph node neighborhood information and generate a dense vector representation of nodes, which can enhance the recommendation effect in the form of incorporating the graph information representation in the deep recommendation model with Transformer as the sequence feature extractor. The proposed method is experimentally validated to shorten the response time and improve the performance of the system, which can increase the user experience of the visual communication system. The system designed in this article is user experience oriented, combined with Multimodal Features and intelligent recommendation algorithms, effectively meeting the personalized needs of users and has certain practical significance.

Author 1: Chao Peng

Keywords: Visual communication system; user experience-oriented; multimodal features; recommendation algorithm

Paper 112: Synchronous Update and Optimization Method for Large-Scale Image 3D Reconstruction Technology Under Cloud-Edge Fusion Architecture

Abstract: Aiming at the problems of limited bandwidth and network delay in the traditional centralized cloud computing mode during large-scale image processing of transmission and distribution digital corridors, a synchronous updating and optimization method for large-scale image 3D reconstruction technology under cloud-edge fusion architecture is proposed. Based on the cloud-side fusion architecture, the image data of the transmission and distribution corridor is preprocessed, feature extraction is performed by deep learning, synchronous updating is performed by using the cloud-side cooperative network, and matching and 3D reconstruction are performed according to the order of the point cloud data; given the dynamically changing characteristics of the image data in the cloud-side fusion environment, the incremental learning is combined with the continuous learning and synchronous updating of the model parameters, to realize the adaptive updating mechanism. The research method utilizes the advantage of cloud-edge fusion architecture to distribute the computational tasks to the cloud and edge, realizing parallel processing and load balancing, and improving the accuracy and efficiency of 3D reconstruction. The experimental results show that the research method in this paper has an image feature point matching rate as high as 96.72%, a lower network latency rate, and a higher real-time performance, which provides strong technical support for the optimization of the transmission and distribution digital corridor 3D reconstruction technology.

Author 1: Jian Zhang

Author 2: Jingbin Luo

Author 3: Yilong Chen

Keywords: Cloud-edge fusion; cloud-edge collaboration; 3D reconstruction; synchronized update

Paper 113: The Application of Anti-Collision Algorithms in University Records Management

Abstract: University records management has grown in importance as a result of the quick growth of big data, artificial intelligence, and other technologies. However, university archives management is prone to data loss, redundancy, and errors. Moreover, the use of scientific management systems and algorithms can effectively improve such problems. To create an effective and secure archive management system and run simulation tests, the study suggests an RFID-based archive management system and uses nested random time slot ALOHA (RS0) and binary tree (BT) anti-collision algorithms to solve the collision problem between tags in the created system. The test results showed that the average query coefficient, recognition efficiency, and communication volume of the proposed algorithm were 1 and 1.2 times, 95% and 90%, 50 Bit and 180 Bit in two scenarios, continuous and uniform, respectively. 0.91% and 3.92%, 24.21% and 31.14% of the system CUP and memory occupation were achieved when the number of clients was 10 and 100, respectively. The average response time of the system was 0.112s and 1.244s when 100 and 1000 users were accessed, respectively. The information extraction accuracy of the system was 94% at 1000 accessed users. This suggests that the approach used in the study can significantly improve the operational effectiveness of the records management system and the accuracy of information extraction, as well as provide technical support for improving the university records management system.

Author 1: Ying Wang

Author 2: Ying Mi

Keywords: Anti-collision algorithms; archive management systems; information networks; RFID technology

Paper 114: Advancements in Deep Learning Architectures for Image Recognition and Semantic Segmentation

Abstract: This paper focuses on using Convolutional Neural Networks (CNNs) for tasks such as image classification. It covers both pre-trained models and those that are built from scratch. The paper begins by demonstrating how to utilize the well-known AlexNet model, which is highly effective for image recognition due to transfer learning. It then explains how to load and prepare the MNIST dataset, a common choice for testing image classification methods. Additionally, it introduces a custom CNN designed specifically for recognizing MNIST digits, outlining its architecture, which includes convolutional layers, activation functions, and fully connected layers for capturing handwritten numbers' details. The paper also guides starting the model, running it on sample data, reviewing outputs, and assessing the accuracy of predictions. Furthermore, it delves into training the custom CNN and evaluating its performance by comparing it with established benchmarks, utilizing loss functions and optimization techniques to fine-tune the model and assess its classification accuracy. This work integrates theory with practical application, serving as a comprehensive guide for creating and evaluating CNNs in image classification, with implications for both research and real-world applications in computer vision.

Author 1: Divya Nimma

Author 2: Arjun Uddagiri

Keywords: Convolutional Neural Networks (CNNs); AlexNet; image classification; transfer learning; MNIST Dataset; Custom CNN Architecture; deep learning; model training and evaluation; neural network optimization; activation functions; feature extraction; machine learning; pattern recognition; data preprocessing; loss functions; model accuracy

Paper 115: Optimized Retrieval and Secured Cloud Storage for Medical Surgery Videos Using Deep Learning

Abstract: Efficient secured storage and retrieval of medical surgical videos are essential for modern healthcare systems. Traditional methods often struggle with scalability, accessibility, and data security, necessitating innovative solutions. This study introduces a novel deep learning-based framework that leverages a hybrid algorithm combining a Variational Autoencoder (VAE) and Group Lasso for optimized video feature selection. This approach reduces dimensionality and enhances the retrieval accuracy of video frames. For storage and retrieval, the system employs a weighted graph-based prefetching algorithm to manage encrypted video data on the cloud, ensuring both speed and security. To ensure data security, video frames are encrypted before cloud storage. Experimental results show that this system outperforms current methods in retrieval speed and accuracy of 99% while maintaining data security. This framework is a significant advancement in medical data management, offering potential applications across other fields that require secure handling of large data volumes.

Author 1: Megala G

Author 2: Swarnalatha P

Keywords: Medical video storage; feature selection; Variational Auto Encoder (VAE); weighted-graph-based prefetching algorithm; group lasso

Paper 116: Rolling Bearing Reliability Prediction Based on Signal Noise Reduction and RHA-MKRVM

Abstract: In order to solve the problem of reliability assessment and prediction of rolling bearings, a noise reduction method (CEEMDAN-GRCMSE) based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN) combined with generalized refined composite multi-scale sample entropy (GRCMSE) is proposed from the vibration signals to remove the noise from the bearing vibration signals, and then the feature set of the noise-reduced signals is downscaled by using the Uniform manifold approximation and projection(UMAP) algorithm, and the reliability assessment model is established by using a logistic regression algorithm to establish a reliability assessment model, and use the red-tailed hawk algorithm for parameter optimization of the mixed kernel relation vector machine, which is used to predict the bearing state, and finally the predicted state information is brought into the assessment model to obtain the final results. In this paper, the whole life cycle data of rolling bearings from Xi ’an Jiaotong University-Sun Science and Technology Joint Laboratory (XJTU-SY) are used to verify the effectiveness of the proposed method. The superiority of the proposed method is highlighted by comparing the analysis results with those of other AI methods.

Author 1: Yifan Yu

Keywords: Rolling bearing; reliability evaluation and pre-diction; complete ensemble empirical mode decomposition with adaptive noise; generalized refined composite multi-scale sample entropy; uniform manifold approximation and projection; red-tailed hawk algorithm; mixed kernel relevance vector machine

Paper 117: EiAiMSPS: Edge Inspired Artificial Intelligence-based Multi Stakeholders Personalized Security Mechanism in iCPS for PCS

Abstract: Artificial Intelligence (AI) is becoming more prevalent in the healthcare sector like in pharmaceutical care to achieve rapid and precise outcomes. Machine learning techniques are critical in preserving this balance since they ensure both the confidentiality and authenticity of healthcare data. Early sickness projections benefit clinicians when establishing early monetary choices, in the lives of their patients. The Web of Things (IoT) is acting as an accelerator to boost the efficacy of AI applications in healthcare. Healthcare service pharmaceutical care is also in demand and can have AI for good patient care. The sensor gathers the data from individuals, then the data is examined employing machine learning algorithms. The work’s major intent is to come up with an automated learning-based user authentication algorithm for providing secure communication. The other goal is to ensure data privacy for sensitive information that does not currently have security. The Federated Learning (FL) technique, which uses a decentralized environment to train models, can be utilized for this purpose. It enhances data privacy. This work proposes in addition to security a differential privacy preservation strategy that involves introducing random noise to a data sample to generate anonymity. The model’s performance and data quality are assessed, as privacy preservation approaches frequently reduce data quality.

Author 1: Swati Devliyal

Author 2: Sachin Sharma

Author 3: Himanshu Rai Goyal

Keywords: Internet of things; pharmaceutical care; machine learning; authentication; artificial intelligence

Paper 118: Integrated IoT-Driven System with Fuzzy Logic and V2X Communication for Real-Time Speed Monitoring and Accident Prevention in Urban Traffic

Abstract: Road safety is a critical concern globally, with speeding being a leading cause of traffic accidents. Leveraging advanced technologies can significantly enhance the ability to monitor and control vehicle speeds in real time. Traditional methods of speed monitoring are often limited in their ability to provide real-time, adaptive interventions. Existing systems do not adequately integrate sensor data and decision-making processes to prevent speeding-related accidents effectively. This paper aims to address these limitations by proposing a novel system that utilizes Internet of Things (IoT) technology combined with fuzzy logic to monitor vehicle speeds and prevent accidents in real time. The proposed system integrates IoT sensors for continuous vehicle speed monitoring and employs a Fuzzy Inference System (FIS) to make decisions based on variables such as speed, alcohol presence, and driver fitness. The system also facilitates interaction between drivers and law enforcement through Vehicle-to-Everything (V2X) communication. The FIS implementation demonstrated effective speed control capabilities, accurately assessing and responding to various risk levels, thereby reducing the likelihood of speeding-related accidents. This research contributes to the advancement of road safety systems by integrating IoT and fuzzy logic technologies, offering a more adaptive and responsive approach to traffic management and accident prevention. Future enhancements will focus on incorporating machine learning techniques to dynamically adjust FIS rules based on real-time data and improve sensor network reliability to ensure more accurate and comprehensive monitoring.

Author 1: Khadiza Tul Kubra

Author 2: Tajim Md. Niamat Ullah Akhund

Author 3: Waleed M. Al-Nuwaiser

Author 4: Md Assaduzzaman

Author 5: Md. Suhag Ali

Author 6: M. Mesbahuddin Sarker

Keywords: Internet of Things; high-speed monitoring; alcohol detection; matlab simulation; write FIS

Paper 119: TGMoE: A Text Guided Mixture-of-Experts Model for Multimodal Sentiment Analysis

Abstract: Multimodal sentiment analysis seeks to determine the sentiment polarity of targets by integrating diverse data types, including text, visual, and audio modalities. However, during the process of multimodal data fusion, existing methods often fail to adequately analyze the sentimental relationships between different modalities and overlook the varying contributions of different modalities to sentiment analysis results. To address this issue, we propose a Text Guided Mixture-of-Experts (TGMoE) Model for Multimodal Sentiment Analysis. Based on the varying contributions of different modalities to sentiment analysis, this model introduces a text guided cross-modal attention mechanism that fuses text separately with visual and audio modalities, leveraging attention to capture interactions between these modalities and effectively enrich the text modality with supplementary information from the visual and audio data. Additionally, by employing a sparsely gated mixture of expert layers, the TGMoE model constructs multiple expert networks to simultaneously learn sentiment information, enhancing the nonlinear representation capability of multimodal features. This approach makes multimodal features more distinguishable concerning sentiment, thereby improving the accuracy of sentiment polarity judgments. The experimental results on the publicly available multimodal sentiment analysis datasets CMU-MOSI and CMU-MOSEI show that the TGMoE model outperforms most existing multimodal sentiment analysis models and can effectively improve the performance of sentiment analysis.

Author 1: Xueliang Zhao

Author 2: Mingyang Wang

Author 3: Yingchun Tan

Author 4: Xianjie Wang

Keywords: Multimodal fusion; sentiment analysis; cross modal; mixture of experts

Paper 120: A Simple and Efficient Approach for Extracting Object Hierarchy in Image Data

Abstract: An object hierarchy in images refers to the structured relationship between objects, where parent objects have one or more child objects. This hierarchical structure is useful in various computer vision applications, such as detecting motorcycle riders without helmets or identifying individuals carrying illegal items in restricted areas. However, extracting object hierarchies from images is challenging without advanced techniques like machine learning or deep learning. In this paper, a simple and efficient method is proposed for extracting object hierarchies in images based on object detection results. This method is implemented in a standalone package compatible with both Python and C++ programming languages. The package generates object hierarchies from detection results by using bounding box overlap to identify parent-child relationships. Experimental results show that the proposed method accurately extracts object hierarchies from images, providing a practical tool to enhance object detection capabilities. The source code for this approach is available at https://github.com/saravit-soeng/HiExtract.

Author 1: Saravit Soeng

Author 2: Vungsovanreach Kong

Author 3: Munirot Thon

Author 4: Wan-Sup Cho

Author 5: Tae-Kyung Kim

Keywords: Object hierarchy; object relationship; object detection; computer vision

Paper 121: Priority-Based Service Provision Using Blockchain, Caching, Reputation and Duplication in Edge-Cloud Environments

Abstract: The integration of Multi-access Edge Computing (MEC) and Dense Small Cell (DSC) infrastructures within 5G and beyond networks marks a substantial leap forward in communication technologies. This convergence is critical for meeting the stringent low latency demands of services delivered to Smart Devices (SDs) through lightweight containers. This paper introduces a novel split-duplicate-cache technique seam-lessly embedded within a secure blockchain-based edge-cloud architecture. Our primary objective is to significantly shorten the service initiation durations in high density conditions of SDs and ENs. This is executed by meticulously gathering, verifying, and combining the most optimal chunk candidates. Concurrently, we ensure that resource allocation for services within targeted ENs is meticulously evaluated for every service request. The system challenges and decisions are modeled then represented as a mixed-integer nonlinear optimization problem. To tackle this intricate problem, three solutions are developed and evaluated: the Brute-Force Search Algorithm (BFS-CDCA) for small-scale environments, the Simulated Annealing-Based Heuristic (SA-CDCA) and the Markov Approximation-Based Solution (MA-CDCA) for complex, high-dimensional environments. A comparative analysis of these methods is conducted in terms of solution quality, computational efficiency, and scalability to assess their performance and identify the most suitable approach for different problem instances.

Author 1: Tarik CHANYOUR

Author 2: Seddiq EL KASMI ALAOUI

Author 3: Mohamed EL GHMARY

Keywords: Multi-access Edge-cloud Computing; container base image chunks; replication; fragmentation; service provision; blockchain; Markov approximation

Paper 122: A Data Augmentation Approach to Sentiment Analysis of MOOC Reviews

Abstract: To address the lack of Chinese online course review corpora for aspect-based sentiment analysis, we pro-pose Semantic Token Augmentation and Replacement (STAR), a semantic-relative distance-based data augmentation method. STAR leverages natural language processing techniques such as word embedding and semantic similarity to extract high-frequency words near aspect terms, learns their word vectors to obtain synonyms and replaces these words to enhance sentence diversity while maintaining semantic consistency. Experiments on a Chinese MOOC dataset show STAR improves Macro-F1 scores by 3.39%-8.18% for LCFS-BERT and 1.66%-8.37% for LCF-BERT compared to baselines. These results demonstrate STAR’s effectiveness in improving the generalization ability of deep learning models for Chinese MOOC sentiment analysis.

Author 1: Guangmin Li

Author 2: Long Zhou

Author 3: Qiang Tong

Author 4: Yi Ding

Author 5: Xiaolin Qi

Author 6: Hang Liu

Keywords: Data augmentation; sentiment analysis; MOOC; natural language processing; deep learning

Paper 123: Design and Implementation of Style-Transfer Operations in a Game Engine

Abstract: The image style transfer operations are a kind of high-level image processing techniques, in which a target image is transformed to show a given style. These kind of operations are typically acquired with modern neural network models. In this paper, we aim to achieve the image style-transfer operations in real time, with the underlying computer games. We can apply the style-transfer operations to the all or part of rendering textures in the existing games, to change the overall feeling and appearance of those games. For a computer game or its underlying game engine, the style-transfer neural network models should be executed so fast to maintain the real-time execution of the original game. Efficient data management is also required to achieve deep learning operations while maintaining overall performance of the game as much as possible. This paper compares several aspects of style-transfer neural network models, and its executions in the game engines. We propose a design and implementation way for the real-time style-transfer operations. The experimental result shows a set of technical points to be considered, while applying neural network models to a game engine. We finally shows that we achieved real-time style-transfer operations, with the Barracuda module in the Unity game engine.

Author 1: Haechan Park

Author 2: Nakhoon Baek

Keywords: Style transfer; neural network models; game engine; rendering textures; real-time operations

Paper 124: Under Sampling Techniques for Handling Unbalanced Data with Various Imbalance Rates: A Comparative Study

Abstract: Unbalanced data sets represent data sets that contain an unequal number of examples for different classes. This dataset represents a problem faced by machine learning tools; as in datasets with high imbalance ratios, false negative rate per-centages will be increased because most classifiers will be affected by the major class. Choosing specific evaluation metrics that are most informative and sampling techniques represent a common way to handle this problem. In this paper, a comparative analysis between four of the most common under-sampling techniques is conducted over datasets with various imbalance rates (IR) range from low to medium to high IR. Decision Tree classifier and twelve imbalanced data sets with various IR are used for evaluating the effects of each technique depending on Recall, F1-measure, gmean, recall for minor class, and F1-measure for minor class evaluation metrics. Results demonstrate that Clusters Centroid outperformed Neighborhood Cleaning Rule (NCL) based on recall for all low IR datasets. For both medium, and high IR datasets NCL, and Random Under Sampling (RUS) outperformed the rest techniques, while Tomek Link has the worst effect.

Author 1: Esraa Abu Elsoud

Author 2: Mohamad Hassan

Author 3: Omar Alidmat

Author 4: Esraa Al Henawi

Author 5: Nawaf Alshdaifat

Author 6: Mosab Igtait

Author 7: Ayman Ghaben

Author 8: Anwar Katrawi

Author 9: Mohmmad Dmour

Keywords: Clusters centroid; decision tree; neighborhood cleaning rule; random under sampling; Tomek Link under sampling; unbalanced datasets

Paper 125: Multiclass Chest Disease Classification Using Deep CNNs with Bayesian Optimization

Abstract: Ever since its outbreak, numerous research studies have been initiated worldwide as an attempt for an accurate and efficient diagnosis of COVID-19. In the recent past, patients suffering from various chronic lung diseases, either passed away due to COVID-19 or Pneumonia. Both of these pulmonary diseases are strongly correlated as they share a common set of symptoms and even for medical professionals, it has been difficult to perform discerned diagnosis for both of these diseases. The dire need of the current scenario is a chest-disease diagnosis framework for accurate, precise, real-time and automatic detection of COVID-19 because of its mass fatality rate. The review of various contemporary and previous research works show that the currently avail-able computer-aided diagnosis systems are insufficient for real-time implementation of COVID-19 prediction due to their long training time, substantial memory requirements and excessive computations. This work proposes an optimized hybrid DNN-ML framework by combining Deep Neural Networks’ (DNNs) models and optimized Machine Learning (ML) classifiers along with an efficacious image preprocessing approach. For feature extraction, Deep learning (DL) models namely GoogleNet, EfficientNetB0, and ResNet50 have been deployed and extracted features have been further fed to Bayesian optimized ML classifiers. The two major contributions of this study are, Edge based Region of Interest (ROI) extraction and use of Bayesian optimization approach for configuring optimal architectures of ML classifiers. With extensive experimentation, it has been observed that the proposed optimized hybrid DNN-ML model with encapsulated image preprocessing techniques performed much better as com-pared to various previously existing ML-DNN models. Based on the promising results obtained from this proposed light weight hybrid framework, it has been concluded that, this model can facilitate radiologists, while functioning as an accurate disease diagnosis and support system for early detection of COVID-19 and Pneumonia.

Author 1: Maneet Kaur Bohmrah

Author 2: Harjot Kaur

Keywords: Deep neural networks; machine learning; Bayesian optimization; image preprocessing; COVID-19; pneumonia

Paper 126: Preprocessing Techniques for Clustering Arabic Text: Challenges and Future Directions

Abstract: Arabic is a complex language for text analysis because of its orthographic features, rich synonyms, and semantic style. Thus, Arabic text must be prepared more carefully in the preprocessing stage for the analyzer to improve the quality of the results. Moreover, many preprocessing steps have been proposed to improve the text analyzer quality by reducing high dimensionality, selecting the proper features to describe the text, and enhancing the process speed. This paper deeply investigates and summarizes the use of Arabic preprocessing techniques in Arabic text in general and focuses in-depth on clustering. Moreover, it focuses on seven preprocesses that are now used to prepare Arabic and provides the available tools for each of them; the seven preprocess are tokenization, normalization, stopword removal, stemming, vectorization, lemmatization, and feature selection. In addition, this paper investigates any work that uses synonyms and semantic techniques for preprocessing to prepare the text or reduce the dimensionality of the clustering algorithm. Therefore, this survey investigated nine techniques for Arabic text preprocessing to identify the challenges in this area. Finally, this study aims to serve as a reference for researchers interested in this area, and ends with potential future research directions.

Author 1: Tahani Almutairi

Author 2: Shireen Saifuddin

Author 3: Reem Alotaibi

Author 4: Shahendah Sarhan

Author 5: Sarah Nassif

Keywords: Arabic preprocessing; Arabic language; survey; clustering; Arabic analysis

Paper 127: Impact of Emojis Exclusion on the Performance of Arabic Sarcasm Detection Models

Abstract: The complex challenge of detecting sarcasm in Arabic speech on social media is exacerbated by the language’s diversity and the nature of sarcastic expressions. There is a significant gap in the capability of existing models to effectively interpret sarcasm in Arabic, necessitating more sophisticated and precise detection methods. In this paper, we investigate the impact of a fundamental preprocessing component on sarcasm detection. While emojis play a crucial role in mitigating the absence of body language and facial expressions in modern communication, their impact on automated text analysis, particularly in sarcasm detection, remains underexplored. We examine the effect of excluding emojis from datasets on the performance of sarcasm detection models in social media content for Arabic, a language with a super-rich vocabulary. This investigation includes the adaptation and enhancement of AraBERT pre-training models by specifically excluding emojis to improve sarcasm detection capabilities. We use AraBERT pre-training to refine the specified models, demonstrating that the removal of emojis can significantly boost the accuracy of sarcasm detection. This approach facilitates a more refined interpretation of language, eliminating the potential confusion introduced by non-textual elements. The evaluated AraBERT models, through the focused strategy of emojis removal, adeptly navigate the complexities of Arabic sarcasm. This study establishes new benchmarks in Arabic natural language processing and offers valuable insights for social media platforms.

Author 1: Ghalyah Aleryani

Author 2: Wael Deabes

Author 3: Khaled Albishre

Author 4: Alaa E. Abdel-Hakim

Keywords: Arabic language; AraBERT; sarcasm detecting; data preprocessing; emojis impact; social media content

Paper 128: A Configurable Framework for High-Performance Graph Storage and Mutation

Abstract: In the realm of graph processing, efficient storage and update mechanisms are crucial due to the large volume of graphs and their dynamic nature. Traditional data structures such as adjacency lists and matrices, while effective in certain scenarios, often suffer from performance trade-offs such as high memory consumption or slow update capabilities. To address these challenges, we introduce CoreGraph, an advanced graph framework designed to optimize both read and update performance. CoreGraph leverages a novel segmentation method and in-place update techniques, along with configurable memory allocators and synchronization mechanisms, to enhance parallel processing and reduce memory consumption. CoreGraph’s update throughput (with up to 20x) and analytics performance exceed those of several state-of-the-art graph structures such as Teseo, GraphOne and LLAMA, while maintaining low memory consumption when the workload includes updates. This paper details the architecture and benefits of CoreGraph, highlighting its practical application in traffic data management where it seamlessly integrates with existing systems providing a scalable and efficient solution for real-world graph data management challenges.

Author 1: Soukaina Firmli

Author 2: Dalila Chiadmi

Author 3: Kawtar Younsi Dahbi

Keywords: Data structures; concurrency; graph processing; graph mutations; high-performance computing; traffic management

Paper 129: Detecting Malware on Windows OS Using AI Classification of Extracted Behavioral Features from Images

Abstract: In this research, using dynamic analysis ten critical features were extracted from malware samples operating in isolated virtual machines. These features included process ID, name, user, CPU usage, network connections, memory usage, and other pertinent parameters. The dataset comprised 50 malware samples and 11 benign programs, providing a data for training and testing the models. Initially, text-based classification methods were employed, utilizing feedforward neural networks (FNN) and recurrent neural networks (RNN). The FNN model achieved an accuracy rate of 56%, while the RNN model demonstrated better performance with an accuracy rate of 68%. These results highlight the potential of neural networks in analyzing and identifying malware based on behavioral patterns. To further explore AI's capabilities in malware detection, the extracted features were transformed into grayscale images. This transformation enabled the application of convolutional neural networks (CNN), which excel at capturing spatial patterns. Two CNN models were developed: a simple model and a more complex model. The simple CNN model, applied to the grayscale images, achieved an accuracy rate of 70.1%. The more complex CNN model, with multiple convolutional and fully connected layers, significantly improved performance, achieving an accuracy rate of 88%. The findings from this research underscore the importance of dynamic analysis. By leveraging both text and image-based classification methods, this study contributes to the development of more robust and accurate malware detection systems. It provides a comprehensive framework for future advancements in cybersecurity, emphasizing the critical role of dynamic analysis in identifying and mitigating threats.

Author 1: Nooraldeen Alhamedi

Author 2: Kang Dongshik

Keywords: Malware analysis; dynamic analysis; image classification; malware behavior extraction; text

Paper 130: The Impact of Virtual Collaboration Tools on 21st-Century Skills, Scientific Process Skills and Scientific Creativity in STEM

Abstract: Virtual collaboration tools have become increasingly important in STEM education, especially after the COVID-19 pandemic. These tools offer many benefits, including developing 21st-century skills and fostering scientific process skills and scientific creativity. However, there are concerns regarding their effectiveness across different genders and regions. This study evaluates the impact of the ExxonMobil Young Engineers (EYE) program, which uses the Zoom application, on enhancing 21st-century skills, scientific process skills, and scientific creativity among secondary school students in Malaysia. The participants primarily consist of 520 secondary school students, with teachers acting as facilitators and professional engineers from ExxonMobil serving as instructors. A pre-test survey was conducted to assess students' initial skill levels. The program consisted of three phases: briefing, breakout room activities, and final reflections. After the program, a post-test survey was conducted to evaluate changes in student skills. Data analysis was analyzed using SPSS software by employing descriptive statistics, MANOVA with Wilks' lambda, one-way ANOVA, and partial eta squared to measure the program's impact and the influence of gender and regional factors. The results showed significant improvements in all three skill areas post-intervention: 21st-century skills, scientific process skills, and scientific creativity. Gender differences were significant for 21st-century skills, while regional differences significantly affected scientific process skills. The EYE program could enhance students' STEM-related skills using virtual collaboration tools like Zoom. However, regional and gender differences highlight the importance of adapting programs to address specific challenges and ensuring equitable opportunities for all students.

Author 1: Nur Atiqah Jalaludin

Author 2: Mohamad Hidir Mhd Salim

Author 3: Mohamad Sattar Rasul

Author 4: Athirah Farhana Muhammad Amin

Author 5: Mohd Aizuddin Saari

Keywords: Virtual collaboration tools; 21st-century skills; scientific process skills; scientific creativity; STEM education

Paper 131: The Impact of E-Commerce Drivers on the Innovativeness in Organizational Practices

Abstract: Innovation in e-commerce practices has revolution-ized the way goods and services are purchased or sold online. This relatively new tool for online transaction provides range of access to wealth of information and knowledge needed to facilitate electronic commerce globally using internet network. The case is not the same in the developing countries where e-commerce innovation is deprived of key the components to drives developing economy. To clearly understand innovation in e-commerce diffusion, 375 quantitative data generated from e-commerce organizations in Libya. Statistically analysis of the key drivers of e-commerce innovations focused on the need for a shift in organizational attitude and knowledge through decision making that are committed to meeting customer’s needs. The inter statistical covariance indicated a strong homogeneity between the drivers of e-commerce with mean value range of 4.09 to 4.82 (58.4 % to 68.8% of responses) indicating that 219 to 258 respondents out of 375 are of the same view. There is strong positive correlation between the drivers of e-commerce innovations except for e-commerce management style that has moderate relation and were statistically significant at 0.00 level. This study clearly explained the main factors of interest that are versatile in providing timely delivery of goods, efficient services and in meeting with e-commerce developmental trend.

Author 1: Abdulghader Abu Reemah A Abdullah

Author 2: Ibrahim Mohamed

Author 3: Nurhizam Safie Mohd Satar

Keywords: E-commerce innovation; e-commerce drivers; performance management; decision making; management style