IJACSA Volume 15 Issue 9 - thesai.org

IJACSA Volume 15 Issue 9

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: Evolving Software Architectures from Monolithic Systems to Resilient Microservices: Best Practices, Challenges and Future Trends

Abstract: Microservice architecture has emerged as a widely adopted methodology in software development, addressing the inherent limitations of traditional monolithic and Service-Oriented Architectures (SOA). This paper examines the evolution of microservices, emphasising their advantages in enhancing flexibility, scalability, and fault tolerance compared to legacy models. Through detailed case studies, it explores how leading companies, such as Netflix and Amazon, have leveraged microservices to optimise resource utilisation and operational adaptability. The study also addresses significant implementation challenges, including ensuring data consistency and managing APIs. Best practices, such as Domain-Driven Design (DDD) and the Saga Pattern, are evaluated with examples from Uber's cross-functional teams and Airbnb's transaction management. This research synthesises these findings into actionable guidelines for organisations transitioning from monolithic architectures, proposing a phased migration approach to mitigate risks and improve operational agility. Furthermore, the paper explores future trends, such as Kubernetes and AIOps, offering insights into the evolving microservices landscape and their potential to improve system scalability and resilience. The scientific contribution of this article lies in the development of practical best practices, providing a structured strategy for organisations seeking to modernise their IT infrastructure.

Author 1: Martin Kaloudis

Keywords: Service-Orientated Architecture; SOA; microservices; monolithic architecture; migration

Paper 2: The Effects of IDS/IPS Placement on Big Data Systems in Geo‑Distributed Wide Area Networks

Abstract: Geographically-distributed wide-area networks (WANs) offer expansive distributed and parallel computing capabilities. This includes the ability to advance Wide-Area Big Data (WABD). As data streaming traverses foreign networks, intrusion detection systems (IDSs) and intrusion prevention systems (IDSs) play an important role in securing information. The authors anticipate that securing WAN network topology with IDSs/IPSs can significantly impact wide-area data streaming performance. In this paper, the researchers develop and implement a geographically distributed big data streaming application using the Python programming language to benchmark IDS/IPS placement in hub-and-spoke, custom-mesh, and full-mesh network topologies. The results of the experiments illustrate that custom-mesh WANs allow IDS/IPS placements that maximize data stream packet transfers while reducing overall WAN latency. Hub-and-spoke network topology produces the lowest combined WAN latency over competing network designs but at the cost of single points of failure within the network. IDS/IPS placement in full-mesh designs is less efficient than custom-mesh yet offers the greatest opportunity for highly available data streams. Testing is limited by specific big data systems, WAN topologies, and IDS/IPS technology.

Author 1: Michael Hart

Author 2: Eric Richardson

Author 3: Rushit Dave

Keywords: Information security; network topology; wide-area big data; wide-area networks; wide-area streaming

Paper 3: STAR, a Universal, Repeatable, Strategic Model of Corporate Innovation for Industry Domination

Abstract: Within an existing organization, internal expertise, staffing, compensation, information systems, and market focus may complicate the introduction of new ideas while culture and aversion to risk may completely derail the organizations’ ability to innovate. The STAR model for corporate innovation provides a theoretical model on how to develop and execute innovative practices to overcome these obstacles and achieve significant market penetration and value. The model is a theoretical framework that empowers organizations of all sizes to construct the necessary structures and advocacy needed to create products, services, and internal processes that enable them to dominate the industry in which they participate. The model also provides the mechanism to support the identification, acceptance, and rapid deployment of relevant new technologies that offer an opportunity to create an unfair advantage, something that is very hard to replicate.

Author 1: Ronald Berman

Author 2: Nicholas Markette

Author 3: Robert Vera

Author 4: Tim Gehle

Keywords: Corporate entrepreneurship; innovation model; market dominance; competitive advantage

Paper 4: Control-Driven Media: A Unifying Model for Consistent, Cross-platform Multimedia Experiences

Abstract: Many media providers offer complementary prod-ucts on different platforms to target a diverse consumer base. Online sports coverage, for instance, may include professionally produced audio and video channels, as well as Web pages and native apps offering live statistics, maps, data visualizations, social commentary and more. Many consumers also engage in parallel usage, setting up streaming products and interactive interfaces on available screens, laptops and handheld devices. This ability to combine products holds great promise, yet, with no coordination, cross-platform user experiences often appear inconsistent and disconnected. We present Control-driven Media (CdM), an extension of the current media model that adds support for coordination and consistency across interfaces, devices, products, and platforms while remaining compatible with existing services, technologies, and workflows. CdM promotes online media control as an independent resource type in multimedia systems. With control as a driving force, CdM offers a highly flexible model, opening up for further innovations in automation, personalization, multi-device support, collaboration and time-driven visualization. Furthermore, CdM bridges the gap between continuous media and Web/native apps, allowing the combined powers of these platforms to be seamlessly exploited as parts of a single, consistent user experience. Extensive research in time-dependent, multi-device, data-driven media experiences supports CdM. In particular, CdM requires a generic and flexible concept for online, timeline-consistent media control, for which a candidate solution (State Trajectory) has recently been published. This paper makes the case for CdM, bringing the significant potential of this model to the attention of research and industry.

Author 1: Ingar M. Arntzen

Author 2: Njal T. Borch

Author 3: Anders Andersen

Keywords: Multi-platform; media control; continuous media; data-driven media; interactive media; orchestrated media

Paper 5: AI-Assisted Academic Writing: A Comparative Study of Student-Crafted and ChatGPT-Enhanced Critiques in Ubiquitous Computing

Abstract: This study examines the impact of Large Language Models (LLMs), such as ChatGPT, on the development of academic critique skills among fourth-year Computer Science undergraduates enrolled in a Ubiquitous Computing course. The research systematically evaluates the differences between student-authored critiques and those revised with the aid of ChatGPT, utilizing established readability metrics such as the Flesch-Kincaid Grade Level, Flesch Reading Ease, and Gunning Fog Index. The findings highlight the potential of AI to enhance readability and analytical depth, while also revealing challenges related to dependency, academic integrity, and algorithmic bias. These results extend implications across learning sciences, pedagogy, and educational technology, providing actionable insights into leveraging AI to augment traditional learning methods and enhance critical thinking and personalized education.

Author 1: Edward R Sykes

Keywords: AI in higher education; AI in academic writing; readability metrics; LLM; ethical considerations

Paper 6: A Natural Language Processing Model for the Development of an Italian-Language Chatbot for Public Administration

Abstract: Natural Language Processing models (NLP) are used in chatbots to understand user input, interpret its meaning, and generate conversational responses to provide immediate and consistent assistance. This reduces problem-solving time and staff workload and increases user satisfaction. There are both rule-based chatbots, which use decision trees and are programmed to answer specific questions, and self-learning chatbots, which can handle more complex conversations through continuous learning about data and user interactions. However, only a few chatbots have been developed specifically for the Italian language. The development of chatbots for Public Administration (PA) in the Italian language presents unique challenges, particularly in creating models that can accurately understand and respond to user queries based on complex, context-specific documents. This paper proposes a novel natural language processing (NLP) model tailored to the Italian language, designed to support the development of an advanced Question Answering (QA) chatbot for PA. The core of the proposed model is based on the BERT (Bidirectional Encoder Representations from Transformers) architecture, enhanced with an encoder/decoder module and a highway network module to improve the filtering and processing of input text. The principal aim of this research is to address the gap in Italian-language NLP models by providing a robust solution capable of handling the intricacies of the Italian language within the context of PA. The model is trained and evaluated using the Italian version of the Stanford Question Answering Dataset (SQuAD-IT). Experimental results demonstrate that the proposed model outperforms existing models such as BIDAF in terms of F1-score and Exact Match (EM), indicating its superior ability to provide precise and accurate answers. The comparative analysis highlights a significant performance improvement, with the proposed model achieving an F1-score of 59.41% and an EM of 46.24%, compared to 49.35% and 38.43%, respectively, for BIDAF. The findings suggest that the proposed model offers substantial benefits in terms of accuracy and efficiency for PA applications.

Author 1: Antonio Piizzi

Author 2: Donatello Vavallo

Author 3: Gaetano Lazzo

Author 4: Saverio Dimola

Author 5: Elvira Zazzera

Keywords: Natural Language Processing; chatbot; BERT; transformer; Italian language

Paper 7: Deep Neural Network and Human-Computer Interaction Technology in the Field of Art Design

Abstract: Traditional art design is usually based on the designer’s intuitive creativity. Limited by individual experience, knowledge and imagination, it is difficult to create more abundant and higher quality works, and the workload is huge, which limits the production efficiency of artworks. Through deep neural networks and human-computer interaction technology, the quality of art design can be improved; the workload and cost of designers can be reduced, and more artistic inspiration and tools can be provided to designers. The main contribution of this paper is to propose the use of a Cycle Generative Adversarial Network (Cycle GAN) to realize the automatic conversion of text to image and provide an immersive art experience through human-computer interaction technology such as virtual reality. In addition, the target audience of this paper is art designers and researchers of human-computer interaction technology, aiming to help them break through the traditional creation mode and lead art design to diversification and avant-garde. The content loss rate of character image conversion in Cycle GAN was reduced by 74.5% compared with that of human image conversion. The average peak signal-to-noise ratio of figure images generated by Cycle GAN was 57.9% higher than that of figure images generated by the artificial method. The character images generated by Cycle GAN reduce content loss and are more realistic. Deep neural networks and human-computer interaction technology can promote the development and progress of art design, break the traditional creative mode and bondage, and lead art to be more diversified and avant-garde.

Author 1: Lan Guo

Author 2: Lisha Luo

Author 3: Weiquan Fan

Keywords: Deep neural network; human-computer interaction; Cycle Generative Adversarial Networks; art design; image generation

Paper 8: Development of Real Time Meteorological Grade Monitoring Stations with AI Analytics

Abstract: Air pollution comes in many forms and the basis of measure is the concentration of particles in the air. The quality of air depends on the quantity of pollution measured by a particle sensor that is accurate down to micron-meter consistencies. The size of the pollutants will be ingested by humans and cause respiratory problems and its effects on health conditions. The research will study the measurement of particles using multiple types of light scattering sensors and reference them to the accuracy of meteorological standards for precision in measurement. The sensors will be subjected to extreme conditions to gauge the repeatability and behavior and also long-term deployment usage. This study is required as when deployed on the field, dust particles will degrade the sensors over time. Early detection of sensor sensitivity and maintenance is therefore considered part of the research. Air particle data is volatile and dynamic over time and with that said, mass deployment of these sensors will give a better measurement of pollution data. However, with more and more data, standard statistics used show a basic level indicator and hence the idea of using machine learning algorithms as part of artificial intelligence (AI) processing is adapted for analyzing and also predicting particle data. There is a foreseeable challenge on this as there is no one machine learning for use only for this and multiple models are considered and gauged with the best accuracy using R2 value as low as 0.75 during the entire research. Lastly, with the seamless Internet of Things sensing architecture, the improved spatial data resolution will be improved and can be used to complement the current pollution measurement data for Malaysia in particular.

Author 1: Adrian Kok Eng Hock

Author 2: Chan Yee Kit

Author 3: Koo Voon Chet

Keywords: Air pollution; air particles; PM2.5; PM10; real time; light scattering sensor; neural networks; AI; machine learning; R2; IoT; WSM

Paper 9: CNN-Based Salient Target Detection Method of UAV Video Reconnaissance Image

Abstract: In order to address the challenges of image complexity, capturing subtle information, fluctuating lighting, and dynamic background interference in drone video reconnaissance, this paper proposes a salient object detection method based on convolutional neural network (CNN). This method first preprocesses the drone video reconnaissance images to remove haze and improve image quality. Subsequently, the Faster R-CNN framework was utilized for detection, where in the Region Proposal Network (RPN) stage, the K-means clustering algorithm was used to generate optimized preset anchor boxes for specific datasets to enhance the accuracy of target candidate regions. The Fast R-CNN classification loss function is used to distinguish salient target regions in reconnaissance images, while the regression loss function precisely adjusts the target bounding boxes to ensure accurate detection of salient targets. In response to the potential failure of Faster R-CNN in extreme situations, this paper innovatively introduces a saliency screening strategy based on similarity analysis to finely screen superpixels, preliminarily locate target positions, and further optimize saliency object detection results. In addition, the use of saturation component enhancement and brightness component dual frequency coefficient enhancement techniques in the HSI color space significantly improves the visual effect of salient target images, enhancing image clarity while preserving the natural and soft colors, effectively improving the visual quality of detection results. The experimental results show that this method exhibits significant advantages of high accuracy and low false detection rate in salient object detection of unmanned aerial vehicle (UAV) video reconnaissance images. Especially in complex scenes, it can still stably and accurately identify targets, significantly improving detection performance.

Author 1: Li Na

Keywords: Regional convolutional neural network; K-means clustering; UAV reconnaissance image; salient target detection; task loss function

Paper 10: Construction of Image Retrieval Module for Ethnic Art Design Products Based on DF-CNN

Abstract: With the increasing interest of consumers in ethnic art, more design products with ethnic art characteristics are being displayed. In order to help users easily retrieve related art products, an image retrieval model that can effectively extract data is proposed. The research method strengthens the depth of data mining through weighted methods, main characteristics and local features in images based on the multi-window combination, and uses the deep forest algorithm to expand the decision path and select information gain nodes. By adjusting the weights of convolutional neural networks, the retrieval ability of the model is enhanced. The gradient problem in the propagation process is optimized using residual modules, and the prominent features of the features are strengthened using a bar attention mechanism to optimize the retrieval ability. The results indicated that the loss function of the research model converged within 20 iterations, and the matching degree of the retrieved images in the testing set reached 91.28% after iterative training. The AUC of the research model was 0.876, indicating that the model had a good performance in image retrieval and classification. The retrieval accuracy of the research model was higher than other methods for image data of different specifications. This indicates that the research model has universality for multi-scale image retrieval, which can provide theoretical support for the development of ethnic art design products.

Author 1: Yaru He

Keywords: Image retrieval; main characteristics; local features; deep forest; convolutional neural network; bar attention mechanism; residual module

Paper 11: Application of Speech Recognition Technology Based on Multimodal Information in Human-Computer Interaction

Abstract: Multimodal human-computer interaction is an important trend in the development of human-computer interaction field. In order to accelerate the technological change of human-computer interaction system, the study firstly fuses Connectionist Temporal Classification algorithm and attention mechanism to design a speech recognition architecture, and then further optimizes the end-to-end architecture of speech recognition by using the improved artificial swarming algorithm, to obtain a speech recognition model suitable for multimodal human-computer interaction system. One of them, Connectionist Temporal Classification, is a machine learning algorithm that deals with sequence-to-sequence problems; and the Attention Mechanism allows the model to process the input data in such a way that it can focus its attention on the relevant parts. The experimental results show that, the hypervolume of the improved swarm algorithm converges to 0.861, which is 0.099 and 0.059 compared to the ant colony and differential evolution algorithms, while the traditional swarm algorithm takes the value of 0.676; the inverse generation distance of the improved swarm algorithm converges to 0.194, while that of the traditional swarm, ant colony, and differential evolution algorithms converge to 0.263, 0.342, and 0.246, respectively. Hypervolume and Inverse Generation Distance Measures the diversity and convergence of the solution set. The speech recognition model takes higher values than the other speech recognition models in the evaluation metrics of accuracy, precision, and recall, and the lowest values of the error rate at the character, word, and sentence levels are respectively 0.037, 0.036 and 0.035, ensuring higher recognition accuracy while weighing the real-time rate. In the multimodal interactive system, the experimental group’s average opinion scores, objective ratings of speech quality, and short-term goal comprehensibility scores, and the overall user experience showed a significant advantage over the control group of the other methods, and the application scores were at a high level. The speech processing technology designed in this study is of great significance for improving the interaction efficiency and user experience, and provides certain references and lessons for the research in the field of human-computer interaction and speech recognition.

Author 1: Yuan Zhang

Keywords: Multimodal information; speech recognition; intelligent optimization algorithm; multimodal human-computer interaction; CTC; attention mechanisms; artificial bee colony algorithms

Paper 12: Application of Fuzzy Decision Support System Based on GNN in Anomaly Detection and Incident Response Service of Intelligent Security

Abstract: This paper introduces a fuzzy decision support system (FDSS) based on a graph neural network (GNN) for anomaly detection and intelligent security. The primary aim is to develop a robust system capable of accurately identifying anomalies and providing timely incident response services. GNNs are utilized to capture the complex relationships and features between nodes in graph data, learning the embedded representation of each node through information transfer and aggregation mechanisms, which encapsulate the structural information of the graph. The FDSS leverages these features to construct a fuzzy rule base and perform fuzzy inference, generating decision suggestions that enhance the system's adaptability and robustness in dealing with uncertain data. The challenges addressed include the need for efficient anomaly detection in large-scale surveillance networks, the requirement for fast response times during emergencies, and the necessity for scalable and adaptable systems. Experimental results demonstrate that the GNN-based FDSS surpasses other methods in terms of anomaly detection accuracy, incident response service efficiency, system processing capacity, and model generalization ability. Compared to traditional statistical methods, machine learning models, and deep learning models, the proposed system maintains high precision and recall rates, processes data more efficiently, and adapts well to new datasets.

Author 1: Tao Chen

Author 2: Xiaoqian Wu

Keywords: GNN; fuzzy decision support system; intelligent security; anomaly detection; incident response service

Paper 13: Using Combined Data Encryption and Trusted Network Methods to Improve the Network Security of the Internet of Things Applications

Abstract: With the integration of big data and artificial intelligence, the Internet of Things has rapidly developed as the foundation for collecting data. The data collected by Internet of Things devices is mostly sensitive information, but limited resources can easily lead to data leakage. Therefore, this study adopts a combination of data encryption and trusted networks to improve the network security of the Internet of Things. This study proposes an Internet of Things network security system based on an improved SM9 encryption algorithm and iTLS security protocol. The system uses a key generation center to generate and distribute keys to complete information encryption and decryption, and identity authentication is carried out through dynamic keys. The results indicated that the total time for key generation, encryption, and decryption based on the SM9-iTLS network security system was 3.63 seconds. The total time for key generation, signature, and signature verification in the system was 3.65 seconds, which is better than other Internet of Things network security systems, and it also had better network resource occupancy and latency than other systems. The Internet of Things network security system based on improved SM9-iTLS can not only improve the security of information transmission among Internet of Things devices but also optimize the efficiency of information transmission. The research results have a certain promoting effect on developing the Internet of Things information security field.

Author 1: Yudan Zhao

Keywords: Security protocols; Encryption; Internet of Things; Resource constraints; SM9

Paper 14: Application and Effectiveness of Improving Retrieval Systems Based on User Understanding in Smart Archive Management Systems

Abstract: In traditional archive management systems, keyword-based retrieval systems often fail to meet users' personalized and precise retrieval needs. To solve this problem, a knowledge graph is first constructed using bidirectional long short-term memory networks and conditional random fields and combined with user understanding-based semantic retrieval to obtain an improved personalized retrieval system. The research results show that the improved personalized retrieval system has significantly better retrieval accuracy and recall rate than traditional retrieval systems. The improved personalized retrieval system has retrieval accuracy rates of 90.24%, 89.65%, 87.52%, 96.33%, and 95.18% for students, civil servants, demobilized soldiers, law enforcement personnel, and retirees, respectively, and recall rates of 89.35%, 91.57%, 89.34%, 97.54%, and 96.63%, respectively. Applying it to the smart archive management system, the accuracy of archive retrieval, personalized recommendation accuracy, response time, and user satisfaction are significantly better than conventional management systems. The improvement and introduction of personalized retrieval systems based on user understanding and knowledge graphs have achieved significant results.

Author 1: Chao Yan

Keywords: Knowledge graph; user understanding; retrieval system; smart archive management; BiLSTM-CRF

Paper 15: Automatic Recognition and Labeling of Knowledge Points in Learning Test Questions Based on Deep-Walk Image Data Mining

Abstract: This paper deeply studies and discusses the application of image data mining technology based on the Deep-Walk algorithm in automatic recognition and annotation of knowledge points in learning test questions. With the rapid development of educational informatization, how to effectively mine and label the knowledge points in learning test questions from image data has become an urgent problem to be solved. In this paper, we introduce a novel approach that integrates graph embedding technology with natural language processing techniques. Initially, we leverage the Deep-Walk algorithm to embed the knowledge points present in the test question images, effectively transforming the high-dimensional image data into a low-dimensional vector representation. This transformation meticulously preserves the intricate structural information while meticulously capturing the subtle semantic nuances embedded within the image data. Subsequently, we undertake a thorough semantic analysis of these vectors, seamlessly integrating natural language processing techniques, to facilitate automated recognition with unparalleled precision. This innovative methodology not only elevates the accuracy of knowledge point recognition to new heights but also achieves semantic annotation of these points, thereby furnishing richer, more insightful data support for subsequent intelligent education applications. Through experimental verification, the proposed method has achieved remarkable results on multiple data sets, which proves its feasibility and effectiveness in practical applications. Furthermore, this paper delves into the expansive potential applications of this methodology in the realm of image data mining, encompassing areas such as online education, intelligent tutoring systems, personalized learning frameworks, and numerous other domains. As we look ahead, we aim to refine the algorithm, enhance recognition accuracy, and uncover additional application scenarios, thereby contributing significantly to the intelligent evolution of the education sector.

Author 1: Ying Chang

Author 2: Qinghua Zhu

Keywords: Deep-walk; image data mining; study test questions; knowledge point recognition

Paper 16: Heart-SecureCloud: A Secure Cloud-Based Hybrid DL System for Diagnosis of Heart Disease Through Transformer-Recurrent Neural Network

Abstract: Cardiovascular disease (CVD) has rapidly increased after COVID-19. Several computerized systems have been developed in the past to diagnose CVD disease. However, the high computing expenses of deep learning (DL) models and the complexity of architectures are significant issues. Therefore, to resolve these issues, an accurate diagnosis of CVD disease is required. This paper proposes a hybrid and secure deep learning (DL) system known as Heart-SecureCloud to predict multiclass heart diseases. To develop this Heart-SecureCloud system, four major stages are makeup such as preprocessing and augmentation, feature extraction and transformation, deep learning and hyperparameter optimization, and cloud security. Advanced signal processing and augmentation technologies are applied to ECG data in the preprocessing and augmentation step to enhance data quality. In the feature extraction and transformation step, adaptive wavelet transforms, and feature scaling are used to extract and convert spectral and temporal data. The DL and hyperparameter optimization step utilize a novel hybrid transformer-recurrent neural network model, which is further optimized for accuracy and efficiency using hyperband-GA. Transfer learning refines pre-trained models using domain-specific data. The unique aspect of the Heart-SecureCloud system is its implementation through a secure cloud, which safeguards medical data with encryption and access control mechanisms. The system's efficacy is demonstrated through testing and evaluation on three publicly available datasets, such as MIT-BIH Arrhythmia MIMIC-III Waveform and PTB-ECG. The Heart-SecureCloud DL architecture achieved impressive results of 98.75% of accuracy, 98.80% of recall, 98.70% of precision, and 98.75% of F1-score. Moreover, the Heart-SecureCloud DL underscores its promise for safe medical diagnostics deployment.

Author 1: Talal Saad Albalawi

Keywords: Heart disease diagnosis; deep learning; cloud computing; feature extraction; data security; hyperparameter optimization; encryption

Paper 17: Pre-Encryption Ransomware Detection (PERD) Taxonomy, and Research Directions: Systematic Literature Review

Abstract: Today’s era is witnessing an alarming surge in ransomware attacks, propelled by the increasingly sophisticated obfuscation tools deployed by cybercriminals to evade conventional antivirus defenses. Therefore, there is a need to better detect and obfuscate viruses. This analysis embarks on a comprehensive exploration of the intricate landscape of ransomware threats, which will become even more problematic in the upcoming era. Attackers may practice new encryption approaches or obfuscation methods to create ransomware that is more difficult to detect and analyze. The damage caused by ransomware ranges from financial losses, at best paid for ransom, to the loss of human life. We presented a Systematic Literature Review and quality analysis of published research papers on the topic. We investigated 30 articles published between the year 2018 to the year 2023(H1). The outline of what has been published thus far is reflected in the 30 papers that were chosen and explained in this article. One of our main conclusions was that machine learning ML-based detection models performed better than others. Additionally, we discovered that only a small number of papers were able to receive excellent ratings based on the standards for quality assessment. To identify past research practices and provide insight into potential future guidelines in the pre-encryption ransomware detection (PERD) space, we summarized and synthesized the existing machine learning studies for this SLR. Future researchers will use this study as a roadmap and assistance to investigate the preexisting literature efficiently and effectively.

Author 1: Mujeeb Ur Rehman Shaikh

Author 2: Mohd Fadzil Hassan

Author 3: Rehan Akbar

Author 4: Rafi Ullah

Author 5: K.S. Savita

Author 6: Ubaid Rehman

Author 7: Jameel Shehu Yalli

Keywords: Cybersecurity; ransomware detection; static and dynamic analysis; machine learning; cyber-attacks; security

Paper 18: Texture Feature and Mel-Spectrogram Analysis for Music Sound Classification

Abstract: The categorization of music has received substantial interest in the management of large-scale databases. However, the sound of music classification (MC) is poorly interesting, making it a big challenge. For this reason, this paper has proposed a new robust combining method based on texture feature with Mel-spectrogram to classify Arabic music sound. A music audio dataset consisting of 404 sound recordings for different four classes of Arabic music sounds has been collected. The collected data became available for free on the Kaggle website. Firstly, music sound is transformed into a Mel spectrogram, and then several texture features are extracted from these Mel spectrogram images. A two-dimensional Haar wavelet is applied to each Mel-spectrogram image, and Local Binary Patterns (LBP), Gray Level Co-occurrence Matrix (GLCM), and Histogram of Oriented Gradient (HOG) are utilized for feature extraction. K-nearest neighbors (KNN), random forest (RF), decision tree (DT), logistic regression (LR), AdaBoost, extreme gradient boosting (XGB), and support vector machine (SVM) classifiers were utilized in a comparative analysis of Machine Learning (ML) algorithms. Two different datasets have been employed in order to evaluate the effectiveness of our approach: the collected dataset that the authors had gathered and the global GTZAN dataset. Our method demonstrates superior performance with a five-fold cross-validation. The experimental findings indicated that the XGB exhibited a high accuracy with an average performance of 97.80% for accuracy, 97.72% for F1-Score, 97.75% for recall, and 97.81% for precision.

Author 1: M. E. ElAlami

Author 2: S. M. K. Tobar

Author 3: S. M. Khater

Author 4: Eman. A. Esmaeil

Keywords: Mel-spectrograms; ML; texture features; MC

Paper 19: DeeplabV3+ Model with CBAM and CSPM Attention Mechanism for Navel Orange Defects Segmentation

Abstract: Accurate defect detection of navel oranges is the key to ensuring the quality of navel oranges and extending their storage life. An improved DeeplabV3+ model integrating attention mechanism is proposed to increase the current low recognition accuracy and slow detection speed of defect detection in navel oranges grading and sorting process. The improved lightweight backbone network HECA-MobileV3 is applied in the DeeplabV3+ model to reduce the amount of computational data and improve the image processing speed. In addition, the Convolutional Block Attention Module (CBAM) and Channel Space Parallel Mechanism CSPM are integrated to the DeeplabV3+ model. ASPP structure is redesigned and the low feature extraction network is optimized to enhance the capture of target edge information and improve the segmentation effect of the model. Experimental results show that the proposed model exhibits a better MIoU and MPA with 89.50% and 94.02%, respectively, while reducing parameters by 49.42M and increasing detection speed by 55.6fps, which are 7.27% and 3.51% higher than the basic model. The results are superior than U-Net, SegNet and PSP-Net semantic segmentation networks. As a results, the proposed method provides better real-time performance, which meets the requirements of industrial production for detection accuracy and speed.

Author 1: Guo Jinmei

Author 2: Wan Nurshazwani Wan Zakaria

Author 3: Wei Bisheng

Author 4: Muhammad Azmi Bin Ayub

Keywords: Navel oranges; defect detection; DeeplabV3+; HECA-MobileNetV3; CBAM attention mechanism; CSPM mechanism

Paper 20: A Framework for Capturing Quality Requirements by Integrating the Requirement Engineering Elements in Agile Software Development Methods

Abstract: The early phase of Agile Software Development (ASD) methods is Requirement Engineering (RE). Quality Requirement (QR) is a type of RE that needs to be captured at the initial development phase to reduce rework, time, and maintenance costs. However, QR is one of the issues mentioned in ASD, namely the need for more capability to elicit, analyze, document, and manage QR. Therefore, this research aims to propose a framework for capturing QR to address QR issues in ASD by integrating RE elements, namely the RE phases, Documentation, Roles, and RE techniques. This research was conducted in four phases: 1) undertaking a theoretical study, 2) conducting an exploratory study to identify the current practices and issues to capture QR in ASD, 3) constructing the framework by using the RE elements, and 4) evaluating the framework by conducting ASD practitioners’ view using questionnaires. The questionnaires were then analyzed using descriptive statistics based on the average mean of each element. The result shows the average mean for all elements (4.25), the average mean of each element for the RE phases (4.36), the documentations (4.11), the roles (4.25), and the RE techniques (4.18). The mean distribution of each element is more than 4 out of 5 indicating that the framework to capture QR is verified. Thus, this framework can be used by ASD practitioners as a guideline to capture QR in ASD methods.

Author 1: Yuli Fitrisia

Author 2: Rosziati Ibrahim

Keywords: Quality requirement; requirement engineering; ASD; framework; ASD practitioners

Paper 21: Data Mining for the Analysis of Student Assessment Results in Engineering by Applying Active Didactic Strategy

Abstract: To make improvements in the teaching-learning process in educational institutions such as universities, it is necessary to analyse the results obtained and recorded from applying Active Didactic Strategies and, based on this, to propose improvements that will help to achieve the Student Outcomes established for the subject in question; the problem to be solved is thus defined, and the results to be obtained from the analysis are relevant for the improvement of student performance. The objective is to analyse the results of the student assessment, the basis for the calculation of which is based on the recording of the qualification achieved through the performance indicators defined for each criterion, of the competencies involved and aligned with the Student Outcomes of the problems proposed to the student, applying various data mining techniques. Data mining is used to treat large amounts and types of data to obtain hidden information and reveal states, patterns and trends; as well as in Education to study the behaviour of students in terms of their performance. The methodology used for the development of the work is based on the Cross-Industry Standard Process for Data Mining methodological model, which is widely used in data mining projects. The results obtained reveal that the Student's t-test and Snedecor's F-test are highly significant, as well as the determination of the lowest performance indicators in order to plan future improvement actions towards better student performance and achieve a high level of learning. Concluding that if the same teaching and learning process is applied the result will be very similar, therefore, the students have finished learning very well.

Author 1: César Baluarte-Araya

Author 2: Oscar Ramirez-Valdez

Author 3: Ernesto Suarez-Lopez

Author 4: Percy Huertas-Niquén

Keywords: Student outcome; problem based learning; assessment; performance indicator; data mining

Paper 22: Compactness-Weighted KNN Classification Algorithm

Abstract: The K-Nearest Neighbor (KNN) algorithm is a widely used classical classification tool, yet enhancing the classification ac-curacy for multi-feature large datasets remains a challenge. The paper introduces a Compactness-Weighted KNN classification algorithm using a weighted Minkowski distance (CKNN) to address this. Due to the variability in sample distribution, a method for deriving feature weights based on compactness is designed. Subsequently, a formula for calculating the weighted Minkowski distance using compactness weights is proposed, forming the basis for developing the CKNN algorithm. Com-parative experimental results on five real-world datasets demonstrate that the CKNN algorithm outperforms eight exist-ing variant KNN algorithms in Accuracy, Precision, Recall, and F1 performance metrics. The test results and sensitivity analysis confirm the CKNN's efficacy in classifying multi-feature da-tasets.

Author 1: Bengting Wan

Author 2: Zhixiang Sheng

Author 3: Wenqiang Zhu

Author 4: Zhiyi Hu

Keywords: K-nearest neighbors; feature weight; Minkowski distance; com-pactness

Paper 23: SIEM and Threat Intelligence: Protecting Applications with Wazuh and TheHive

Abstract: The consequences of cyberattacks on enterprises are highly varied. DDoS assaults can render an organization's website inaccessible; SQL attacks can compromise the integrity of data in a database, and Brute Force attacks can lead to unauthorized users gaining control over a server or application. Hence, it is crucial for enterprises to be aware of these potential dangers and employ solutions capable of monitoring networks, apps, and servers. In this study, the author employs Wazuh, TheHive, Telegram, and CVSS. Wazuh functions as a tool for monitoring applications and identifying potential security risks. TheHive classifies threats according to their level of importance. Telegram is utilized for dispatching notifications to the administrator. The findings indicate that Wazuh can promptly identify security risks by verifying that the date and time configurations on each utilized server align with the Indonesian time standard. Several vulnerabilities in the applications were successfully detected. The Wazuh server monitors two specific apps, namely Kompetensi and ESPPD. Surveillance commenced on March 20, 2024, at 17:49 and concluded on June 20, 2024, at 01:10, effectively amassing a total of 16,580 logs. 11 essential alert categories require follow-up due to their potential to compromise the system's integrity, confidentiality, and availability. To validate the detection results, the Common Vulnerability Scoring System (CVSS) is used. The assessment of vulnerability levels varies depending on the Wazuh level and CVSS. This arises because CVSS assigns scores based on five exploitability characteristics and incorporates the expertise of specialists to determine the assessment category and evaluate the potential impact of a successful threat. The outcome of this assessment, involving professional expertise, is heavily influenced by the unique attributes of each company. As a result, even when evaluating the same threats, the assessment can yield varying results. Evaluations utilizing Wazuh and CVSS are highly efficient in determining the extent of discovered hazards. By integrating these two technologies, the produced findings become more accurate.

Author 1: Jumiaty

Author 2: Benfano Soewito

Keywords: Application server security; application vulnerability; threat detection; SIEM; Wazuh; TheHive; Telegram and CVSS

Paper 24: Dynamic Priority-Based Round Robin: An Advanced Load Balancing Technique for Cloud Computing

Abstract: An imbalance of load is an essential problem in cloud computing where the division of work between virtual machines is not well-optimized. Performance bottlenecks result from this unequal resource allocation, which keeps the system from operating at its full capability. Managing this load-balancing issue becomes critical to improving overall efficiency, resource utilization, and responsiveness as cloud infrastructures strive to respond to changing workloads and scale dynamically. Crossing the load-balancing landscape introduced a new strategy to effectively improve the load-balancing factor and ways to improve load-balancing performance by understanding how existing algorithms work, an effective method of load balancing. The "Dynamic Priority Based Round Robin" algorithm is a new approach that combines three different algorithms to improve cloud load balancing. This method improves load balancing by taking the best aspects of previous algorithms and improving them. It works remarkably well and responds quickly to commands, greatly reducing processing time. This DPBRR algorithm also plays an important role in improving cloud load balancing in many ways, including improving resource consumption, inefficiency, scalability, fault tolerance, cost optimization, and other aspects. Since it is a combination of algorithms, it may have its drawbacks, but its cloud computing enhancements are very useful for doing many tasks quickly. Strength and adaptability are quite effective, as is adaptability.

Author 1: Parupally Venu

Author 2: Pachipala Yellamma

Author 3: Yama Rupesh

Author 4: Yerrapothu Teja Naga Eswar

Author 5: Maruboina Mahiddar Reddy

Keywords: Load balancer; traffic distribution; cloud computing; resource utilization; scalability; Dynamic Priority Based Round Robin (DPBRR)

Paper 25: Enhancing BLDC Motor Speed Control by Mitigating Bias with a Variation Model Filter

Abstract: Brushless DC motors (BLDC) are integral to a wide array of applications, from electric vehicles to industrial machinery, due to their superior efficiency, reliability, and performance. Effective control of BLDC motors is essential to leverage their full potential and ensure optimal operation. Traditional PID controllers often fall short in handling the nonlinear and dynamic characteristics of BLDC systems, while advanced methods like Active Disturbance Rejection Control (ADRC) introduce additional complexity and cost. This research proposes a Variation Model Filter (VMF) based control system that estimates and compensates for the total bias arising from parameter variations and internal uncertainties. This method simplifies the control process, enhances robustness, and boosts performance without requiring extensive parameter tuning or high costs. Additionally, the paper provides a comprehensive mathematical model for the speed dynamics of BLDC motors. Simulation results based on MATLAB/Simulink indicate that the VMF-based PID control system surpasses both linear ADRC and traditional PID controllers in managing speed dynamics and responding to load disturbances. This approach offers an efficient and cost-effective solution for BLDC motor speed control, with significant potential for broader application and further optimization in motor control systems.

Author 1: Abdul Rahman Abdul Majid

Keywords: EV’s motors; brushless direct current (BLDC) motor; active disturbances rejection control (ADRC); disturbance rejection; bias estimation; Variation Model Filter (VMF)

Paper 26: Dynamic Monitoring of Bridge Structures via an Integrated Cloud and Edge Computing System

Abstract: Traditional bridge monitoring techniques, which predominantly rely on centralized data processing, often exhibit slow and inflexible responses when managing large-scale sensor network data. This study proposes an integrated edge and cloud computing approach to enhance the response time and data processing efficiency of dynamic bridge structure monitoring systems, thereby improving bridge safety and reliability. The proposed monitoring system leverages both edge and cloud computing, incorporating modules such as sensor data management, structural assessment and warning, data processing, monitoring, and data acquisition and transmission. High-performance and cost-effective sensors are utilized to monitor the real-time dynamic responses of the bridge, including displacement, acceleration, tilt, and stress, as well as external loads and environmental effects. The data processing module employs the modal superposition method, frequency response function, and modal analysis for dynamic analysis, while the cloud computing platform facilitates deep learning analysis and long-term data storage. A real case study demonstrates the system's performance across various settings and operational conditions, highlighting the effectiveness of integrating edge and cloud computing. The results indicate that the integration scheme significantly enhances monitoring accuracy, system stability, real-time response capacity, and data processing efficiency.

Author 1: Guoqi Zhang

Author 2: Pengcheng Zhang

Author 3: Xingwang Li

Author 4: Yizhe Yang

Keywords: Dynamic monitoring of bridge structures; edge computing; cloud computing; data processing; modal analysis

Paper 27: Optimized Fertilizer Dispensing for Sustainable Agriculture Through Secured IoT-Blockchain Framework

Abstract: Precision farming is essential for optimizing resource use and improving crop yields to attain sustainable agriculture. However, challenges like data insecurity, fertilizer costs, and inadequate consideration of soil health pose a hindrance to achieving these goals. To overcome these issues, the proposed work presents a novel approach for optimizing fertilizer dispensing by developing a framework connecting IoT and blockchain with a community of greenhouses. The system consists of IoT sensors installed inside the greenhouses to measure soil pH and nutrient values. This collected sensor data is compressed and stored securely and in an off-chain manner by the IPFS (Inter-Planetary File System) hash using the Keccak-256. MetaMask transfers the data for blockchain registration and authentication. The data is then preprocessed using Z-score normalization, Label Encoding, and One-Hot Encoding to obtain a precise analysis. A Deep Learning-based Convolutional Neural Network (DL-CNN) is used to classify soil conditions and determine the appropriate fertilizer requirements. The results of the DL-CNN model are viewed in a dashboard through a Decentralized Application (D-App) that we developed to provide real-time information to consumers, field analysts, and agricultural organizations. Field analysts use the information to establish a control center for precisely applying fertilizers. The proposed method achieves a classification accuracy rate of 98.86%, thus increasing soil health and providing a solution for effectively managing fertilizers.

Author 1: B. C. Preethi

Author 2: G. Sugitha

Author 3: T. B. Sivakumar

Keywords: Fertilizer dispensing; IoT sensors; blockchain; deep learning; convolutional neural network; greenhouse management; and decentralized application

Paper 28: A Capacity-Influenced Approach to Find Better Initial Solution in Transportation Problems

Abstract: Finding an Initial Basic Feasible Solution (IBFS) is the first and essential step in obtaining the optimal solution for any Transportation Problem. Numerous approaches are available in the literature to determine the IBFS; however, many of these methods are modifications of Vogel's Approximate Method (VAM) and/or the Least Cost Method (LCM). None of the existing methods directly consider the capacity of distributions among the nodes when selecting the allocation steps. While researchers have proposed various approaches and demonstrated improved solutions with numerical instances, they have not thoroughly investigated the underlying causes of these results. In this article, we explore the impact of capacity distributions among the nodes on the VAM and LCM in an experimental domain. The study introduces a novel and unique Capacity-Influenced Distribution Indicator (CI-DI) designed to control the flow of allocation. Ultimately, we propose a novel Capacity-Influenced approach that embeds both LCM and VAM to determine the IBFS for Transportation Problems (TPs). The novelty of the proposed approach lies in its direct consideration of capacity distribution among the nodes in the flow of allocations, this feature is lacking in LCM, VAM, and other established approaches. The proposed method develops a novel distribution indicator and a novel cost entry embedded capacity-based matrix to control the flow of allocations and thereby finds the IBFS for the Transportation Problem. We have conducted extensive numerical experiments to assess the effectiveness of the proposed approach. Experimental analysis demonstrates that the proposed method is more efficient in finding the IBFS than existing approaches. Moreover, as it uses a one-time generated Distribution Indicator (DI) for all steps of allocation, it is computationally cheaper than VAM, which generates a DI for each step of allocation.

Author 1: Md. Toufiqur Rahman

Author 2: A R M Jalal Uddin Jamali

Author 3: Momta Hena

Author 4: Mohammad Mehedi Hassan

Author 5: Md Rafiul Hassan

Keywords: Transportation problem; least cost method; Vogel’s approximate method; cost matrix; transportation tableau; node; capacity; route; capacity-influenced; weighted opportunity cost

Paper 29: RSS-LSTM: A Metaheuristic-Driven Optimization Approach for Efficient Text Classification

Abstract: The digital data consumed by the average user daily is huge now and is increasing daily all over the world, which requires sophisticated methods to automatically process data, such as retrieving, searching, and formatting the data, particularly for classifying text data. Long Short-Term Memory (LSTM) is a prominent deep learning model for text classification. Several metaheuristic approaches, such as the Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Firefly Algorithm (FF), have also been used to optimize Deep Learning (DL) models for classification. This study introduced an improved technique for text classification, called RSS-LSTM. The proposed technique optimized the hyperparameters and kernel function of LSTM through the Ringed Seal Search (RSS) algorithm to enhance simplification and learning ability. This work was also compared and evaluated against state-of-the-art techniques such as GA-LSTM, PSO-LSTM, and FF-LSTM. The results showed significantly better results using the proposed techniques, with an accuracy of 96%, recall of 96%, precision of 96%, and 95% f-measure on the Reuters-21578 dataset. In addition, it showed an accuracy of 77%, recall of 77%, precision of 78%, and f-measure of 76% on the 20 Newsgroups dataset, while it achieved accuracy, recall, precision, and f-measure of 91%, 91%, 94%, and 90%, respectively, using the AG News dataset.

Author 1: Muhammad Nasir

Author 2: Noor Azah Samsudin

Author 3: Shamsul Kamal Ahmad Khalid

Author 4: Souad Baowidan

Author 5: Humaira Arshad

Author 6: Wareesa Sharif

Keywords: Deep learning; text classification; Long Short-Term Memory; Ringed Seal Search; metaheuristic algorithms; Part Swarm Optimization; Genetic Algorithm; Firefly Algorithm; hyperparameter optimization

Paper 30: Construction of Image Retrieval Module for Cultural and Creative Products Based on DF-CNN

Abstract: With the growth of the cultural and creative product industry, more and more cultural and creative products have been designed and published in different channels. A method based on image retrieval module is proposed to address the search problem of Chinese creative products in online channels. During the process, a cascaded forest is proposed to achieve layer by layer processing, with class vectors as the main transfer content in the entire forest system. An image attribute feature extraction process that introduces extreme gradient enhancement is designed, and the aggregation of multi-scale and multi-region features is utilized to improve image retrieval performance. The experimental results showed that in the similarity test of extracting image feature information when the image contained three composite cultural and creative objects and the total pixel amount of the image reached 7M, the similarity of image feature information was 97.6%. In the analysis of running time, the research method only took 7.4ms to generate search results in seven fields. In the analysis of the proportion of false search content, the research method maintained a false search proportion of within 6.0% when searching for a single cultural and creative product object. This indicates that the research method has higher accuracy and efficiency in image retrieval of cultural and creative products. Research methods can provide certain technical support for the development of the cultural and creative industry.

Author 1: Meng Jiang

Keywords: Image retrieval; class vector; extreme gradient enhancement; Chinese creative products; layer by layer processing

Paper 31: Method for Mission Analysis Using ToT-Based Prompt Technology Utilized Generative AI

Abstract: Method for mission analysis using ToT: Tree of Thought-based prompt technology utilized generative analysis AI is proposed. Mission analysis needs methods for simulation of the supposed images which will be acquired with the imaging mission instruments, and the other mission instruments. In order to create simulation images, ToT-based prompt technology utilized generative AI is used. An application of the proposed method is shown for a mission analysis for SaganSat-0 of remote sensing satellite which will carry three mission instruments, a 720-degree camera, a thermal infrared camera and a Geiger counter. The simulated images and the Geiger counter sounds created by the proposed method are shown here together with analyzed results.

Author 1: Kohei Arai

Keywords: CoT; ToT; AI; Mission analysis; prompt technology; generative AI; SaganSat-0; remote sensing satellite; 720-degree camera; thermal infrared camera; Geiger counter

Paper 32: IoT-Based Integrated Heads-Up Display for Motorcycle Helmet

Abstract: The prevalence of visual impairment among the global population is a growing concern, with rates continuing to rise at an alarming pace. According to statistics from the World Health Organization (WHO), an estimated 2.2 billion people globally live with some form of visual impairment. Several methods exist to aid the blind in everyday navigation, such as walking sticks and guide dogs. However, these aids do not come without their drawbacks. For instance, using traditional guide dogs may not be suitable for some individuals due to allergies, cultural beliefs, or being unable to take care of a living animal due to the level of responsibility required. Innovations such as smart walking sticks and robotic guide dogs are continually being developed to overcome these gaps and cater to the unique requirements of the visually impaired. Hence, this proposed system is equipped with a joystick-controlled robotic guide that mimics the responsibilities of a traditional guide dog. The proposed system features an obstacle avoidance feature that will detect obstacles in its environment to avoid collisions. It will also provide audio feedback through a Bluetooth-connected mobile application when an obstacle has been detected. The proposed system is a product innovation which can be targeted to benefit visually impaired users by providing them with more independence as well as convenience in terms of mobility. Upon performing acceptance testing with the target audience, the system has been found to achieve its target in aiding the guidance of blind individuals.

Author 1: L. Raj

Author 2: M. Batumalay

Author 3: C. Batumalai

Author 4: Prabadevi B

Keywords: Heads-up display; motorcycle helmet; Internet of Things (IoT); android application; Raspberry Pi; product innovation

Paper 33: A Robust Wrapper-Based Feature Selection Technique Using Real-Valued Triangulation Topology Aggregation Optimizer

Abstract: Feature selection is a critical preprocessing technique used to remove irrelevant and redundant features from datasets while maintaining or improving the accuracy of machine learning models. Recent advancements in this area have primarily focused on wrapper-based feature selection methods, which leverage metaheuristic search algorithms (MSAs) to identify optimal feature subsets. In this paper, we propose a novel wrapper-based feature selection method utilizing the Triangulation Topology Aggregation Optimizer (TTAO), a newly developed algorithm inspired by the geometric properties of triangular topology and similarity. To adapt the TTAO for binary feature selection tasks, we introduce a conversion mechanism that transforms continuous decision variables into binary space, allowing the TTAO—which is inherently designed for real-valued problems—to function efficiently in binary domains. TTAO incorporates two distinct search strategies, generic aggregation and local aggregation, to maintain an effective balance between global exploration and local exploitation. Through extensive experimental evaluations on a wide range of benchmark datasets, TTAO demonstrates superior performance over conventional MSAs in feature selection tasks. The results highlight TTAO's capability to enhance model accuracy and computational efficiency, positioning it as a promising tool to advance feature selection and support industrial innovation in data-driven tasks.

Author 1: Li Pan

Author 2: Wy-Liang Cheng

Author 3: Sew Sun Tiang

Author 4: Kim Soon Chong

Author 5: Chin Hong Wong

Author 6: Abhishek Sharma

Author 7: Touseef Sadiq

Author 8: Aasam Karim

Author 9: Wei Hong Lim

Keywords: Classification; exploration; exploitation; feature selection; metaheuristic search algorithm; machine learning; optimization; triangulation topology aggregation optimizer

Paper 34: Mixed Integer Programming Model Based on Data Algorithms in Sustainable Supply Chain Management

Abstract: With the deepening of globalization and increasing demands for environmental sustainability, modern supply chains are faced with increasingly complex management challenges. To reduce management costs and enhance efficiency, an experimental approach is proposed based on a Mixed Integer Programming Model, integrating heuristic algorithms with adaptive genetic algorithms. The objective is to improve both the efficiency and sustainability of supply chain management. Initially, the selection of suppliers within the supply chain is analyzed. Subsequently, heuristic algorithms and genetic algorithms are jointly employed to design, generate, and optimize initial solutions. Results indicate that during initial runs on training and validation sets, the fitness values of the research method reached as high as 99.67 and 96.77 at the 22nd and 68th iterations, respectively. Moreover, on the training set with a dataset size of 112, the accuracy of the research method was 98.56%, significantly outperforming other algorithms. With the system running five times, the time consumed for supplier selection and successful order allocation was merely 0.654s and 0.643s, respectively. In practical application analysis, when the system iterated 99 times, the research method incurred the minimum total cost of 962,700 yuan. These findings demonstrate that the research method effectively minimizes supply chain management costs while maximizing efficiency, offering practical strategies for optimizing and sustainably developing supply chain management.

Author 1: Shaobin Dong

Author 2: Aihua Li

Keywords: Mixed Integer Programming Model; sustainable; supply chain management; heuristic algorithm; adaptive genetic algorithm

Paper 35: DBRF: Random Forest Optimization Algorithm Based on DBSCAN

Abstract: The correlation and redundancy of features will directly affect the quality of randomly selected features, weakening the convergence of random forests (RF) and reducing the performance of random forest models. This paper introduces an improved random forest algorithm—A Random Forest Algorithm Based on DBSCAN (DBRF). The algorithm utilizes the DBSCAN (Density-Based Spatial Clustering of Applications with Noise) algorithm to improve the feature extraction process, to extract a more efficient feature set. The algorithm first uses DBSCAN to group all features based on their relevance and then selects features from each group in proportion to construct a feature subset for each decision tree, repeating this process until the random forest is built. The algorithm ensures the diversity of features in the random forest while eliminating the correlation and redundancy among features to some extent, thereby improving the quality of random feature selection. In the experimental verification, the classification prediction results of CART, RF, and DBRF, three different classifiers, were compared through ten-fold cross-validation on six different-sized datasets using accuracy, precision, recall, F1, and running time as validation indicators. Through experimental verification, it was found that DBRF algorithm outperformed RF, and the prediction performance was improved, especially in terms of time complexity. This algorithm is suitable for various fields and can effectively improve the classification prediction performance at a lower complexity level.

Author 1: Wang Zhuo

Author 2: Azlin Ahmad

Keywords: Random forest; DBSCAN; feature selection; feature redundancy; classification algorithm

Paper 36: Enhancing Emergency Response: A Smart Ambulance System Using Game-Building Theory and Real-Time Optimization

Abstract: Dispatching ambulances early and efficiently is paramount and difficult in the field of emergency medical services. In this regard, the paper designs a smart ambulance system based on game-building theory. The system employs an advanced Negamax algorithm for optimizing the dispatch of ambulances during emergencies. Besides traditional methods, real-time traffic data, patient condition severity, and dynamic resource allocation also improve the system further. With the integration of predictive analytics and real-time data, it allows dynamic adaptation to changing urban conditions, optimal resource allocation as well as minimizing response time. According to our simulations involving extensive scenarios, our Negamax-based system performs significantly better with respect to average response times when compared with traditional methods averagely reducing them by more than 50%, hence, showing double improvement. The study not only improves efficiency in the operation of emergency services but also presents an expandable framework that can be used for future developments in critical response systems thereby leading to their association with smart city infrastructure and AI-based predictive emergency management.

Author 1: Guneet Singh Bhatia

Author 2: Azhar Hussain Mozumder

Author 3: Saied Pirasteh

Author 4: Satinder Singh

Author 5: Moin Hasan

Keywords: Emergency medical services; ambulance dispatch optimization; advanced game theory; Negamax algorithm; real-time optimization; predictive analytics

Paper 37: Deep Reinforcement Learning-Based Carrier Tuning Algorithm for Mobile Communication Networks

Abstract: With the evolution of mobile communication networks towards 5G and beyond to 6G, managing network resources presents unprecedented challenges, particularly in scenarios demanding high data rates, low latency, and extensive connectivity. Traditional resource allocation methods struggle with network dynamics and complexity, including user mobility, varying network loads, and diverse Quality of Service (QoS) requirements. Deep Reinforcement Learning (DRL), an emerging AI technique, demonstrates significant potential due to its adaptive and learning capabilities. This paper integrates user mobility and network load prediction into a DRL framework and proposes a novel reward function to enhance resource utilization efficiency while meeting real-time QoS demands. We establish a system model involving base stations and receiving terminals to simulate communication services within coverage areas. Comparative experiments analyze the performance of the DRL approach versus traditional methods across metrics such as throughput, delay, and spectral efficiency. Results indicate DRL's superiority in handling dynamic environments and fulfilling QoS needs, especially under heavy loads. This study introduces innovative approaches and tools for future mobile network resource management, paving the way for practical DRL implementations and enhancing overall network performance.

Author 1: Weimin Zhang

Author 2: Xinying Zhao

Keywords: DRL; mobile network; carrier tuning

Paper 38: Human-Computer Interaction Standardization and Systematization Development

Abstract: Amidst the information technology boom, this study harnesses IT and human-computer interaction to revolutionize English education. Our model, grounded in literature review and fieldwork, implemented teaching experiments that enhanced students' English proficiency by 25%, particularly in listening and speaking. Student engagement and interest surged by 30%, underscoring the effectiveness of standardized, systematized English education in the digital era. The study advocates for broader adoption of informatization in teaching, emphasizing the pivotal role of teachers in facilitating this educational shift. With significant outcomes, our research paves the way for future enhancements in English education, ensuring quality and equity in learning. Our approach addresses the gap between traditional teaching and technological advancements, offering personalized learning experiences that improve student outcomes. It also ensures consistent teaching quality and bridges educational divides. We introduce an informatization-based English education model, supported by literature review, fieldwork, and teaching experiments. Our findings show significant improvements in students' English proficiency, highlighting the model's effectiveness.

Author 1: Xiaoling Lyu

Author 2: Hongmiao Yuan

Author 3: Zhao Zhang

Keywords: Human-computer interaction; information technology; English education; standardization; systematization

Paper 39: Eavesdropping Interference in Wireless Communication Networks Based on Physical Layer Security

Abstract: Effective communication security protection can protect people's privacy from being violated. To raise the communication security of wireless communication networks, a collaborative eavesdropping interference scheme with added artificial noise is proposed by combining physical layer security and clustering scenarios to protect the communication security of wireless sensor networks. This scheme adds artificial noise to the transmitted signal to interfere with the eavesdropping signal, making the main channel the dominant channel and achieving eavesdropping interference in wireless communication networks. The results show that after using artificial noise, as the signal-to-noise ratio of the main channel increases from 0 to 20dB, the confidentiality capacity can increase from 0.5 to over 4.0. When the transmission power is 0.4W, the confidentiality capacity reaches its maximum and does not depend on the signal-to-noise ratio. When the number of interfering nodes increases from 1 to 2, the confidentiality capacity increases from approximately 4.7 to around 5.8. The research designed a wireless communication network eavesdropping interference scheme that can effectively protect the information security of the wireless communication network, making the main channel an advantageous channel and achieving complete confidentiality. This scheme can be applied to wireless communication networks to improve the security level of the network.

Author 1: Mingming Chen

Author 2: Yuzhi Chen

Keywords: Wireless communication; sensors; network security; eavesdropping interference; clustering scenario

Paper 40: Development of a Hybrid Quantum Key Distribution Concept for Multi-User Networks

Abstract: This paper investigates the increasing concerns related to the vulnerability of contemporary security solutions in the face of quantum-based attacks, which pose significant challenges to existing cryptographic methods. Most current Quantum Key Distribution (QKD) protocols are designed with a focus on point-to-point communication, limiting their application in broader network environments where multiple users need to exchange information securely. To address this limitation, a thorough analysis of twin-field-based algorithms is conducted, emphasizing their distinct characteristics and evaluating their performance in practical scenarios in Sections II, III, and IV. By synthesizing insights from these analyses, integrating cutting-edge advancements in Quantum Communication technologies, and drawing on proven methodologies from established point-to-point protocols, this study introduces a novel concept for a Hybrid Twin-Field QKD protocol in Section IV. This network-oriented approach is designed to facilitate secure communication in networks involving multiple users, offering a practical and scalable solution. The proposed protocol aims to reduce resource consumption while maintaining high-security standards, thereby making it a viable option for real-world quantum communication networks. This work contributes to the development of more resilient and efficient quantum networks capable of withstanding future quantum-based threats.

Author 1: Begimbayeva Y

Author 2: Zhaxalykov T

Author 3: Makarov M

Author 4: Ussatova O

Author 5: Tynymbayev S

Author 6: Temirbekova Zh

Keywords: Quantum key distribution; quantum communication; multi-user networks; network security; quantum-based attacks; cryptography; point-to-point protocols; resource efficiency; cryptography; information security

Paper 41: The Role of Artificial Intelligence in Enhancing Business Intelligence Capabilities for E-Commerce Platforms

Abstract: This research focuses on the application of BERT (Bidirectional Encoder Representations from Transformers) and Graph Neural Networks (GNNs) to improve business intelligence (BI) capabilities on e-commerce platforms. The main aim of the research is to develop automation methods for the classification of customer interactions and to create a more effective product recommendation system. In this study, BERT was used to analyze and classify customer interaction texts, including questions, complaints, and reviews, with accuracy reaching 97% and sentiment analysis accuracy of 93%. GNNs are applied to model complex relationships between customers and products based on transaction data, then used to provide product recommendations. The evaluation results show that the GNNs model achieved a mean average precision (MAP) of 0.92 and a normalized discounted cumulative gain (NDCG) of 0.88, indicating high relevance and accuracy in product recommendations. This research concludes that the integration of BERT and GNNs improves operational efficiency through classification automation but also provides added value in marketing strategies with better personalization of recommendations.

Author 1: Sinek Mehuli Br Perangin-Angin

Author 2: David Jumpa Malem Sembiring

Author 3: Asprina Br Surbakti

Author 4: Soleh Darmansyah

Keywords: Bidirectional Encoder Representations from Transformers (BERT); Graph Neural Networks (GNNs); business intelligence; e-commerce; product recommendation

Paper 42: Elevating Grape Detection Precision and Efficiency with a Novel Deep Learning Model

Abstract: In the domain of modern agricultural automation, precise grape detection in orchards is pivotal for efficient harvesting operations. This study introduces the Grapes Enhanced Feature Detection Network (GEFDNet), leveraging deep learning and convolutional neural networks (CNN) to enhance target detection capabilities specifically for grape detection in orchard environments. GEFDNet integrates an innovative Enhanced Feature Fusion Module (EFFM) into an advanced YOLO architecture, employing a 16x downsampling Backbone for feature extraction. This approach significantly reduces computational complexity while capturing rich spatial hierarchies and accelerating model inference, which is crucial for real-time object detection. Additionally, an optimized dual-path detection structure with an attention mechanism in the Neck enhances the model's focus on targets and robustness against dense grape detection and complex background interference, a common challenge in computer vision applications. Experimental results demonstrate that GEFDNet achieves at least a 3.5% improvement in mean Average Precision (mAP@0.5), reaching 89.4%. It also has a 9.24% reduction in parameters and a 10.35 FPS increase in frame rate compared to YOLOv9. This advancement maintains high precision while improving operational efficiency, offering a promising solution for the development of automated harvesting technologies. The study is publicly available at: https://github.com/YangxuWangamI/GEFDNet.

Author 1: Xiaoli Geng

Author 2: Yaru Huang

Author 3: Yangxu Wang

Keywords: Computer vision; deep learning; Convolutional Neural Networks (CNN); real-time object detection; dual-path detection structure

Paper 43: Vehicular Traffic Congestion Detection System and Improved Energy-Aware Cost Effective Task Scheduling Approach for Multi-Objective Optimization on Cloud Fog Network

Abstract: A current research area called fog computing aims to extend the advantages of cloud computing to network edges. Task scheduling is a crucial problem for fog device data processing since a lot of data from the sensor or Internet of Things layer is generated at the fog layer. This research suggested a vehicular traffic congestion detection model and an energy-aware cost effective task scheduling (ECTS) method in a cloud fog scenario. This research proposes an ECTS approach to allocate jobs to the fog nodes effectively. The recommended scheduling approach minimizes energy consumption and decreases expenses for time-sensitive real-time applications. The ECTS algorithm is implemented, and results are analysed using the iFogSim simulator. The proposed method minimizes energy consumption and cost. The suggested ECTS method is tested with five sets of inputs in this paper. The experiment's results show that an ECTS minimizes energy consumption in comparison to alternative algorithms. It also reduces the execution cost. The suggested approach outperforms both the Round-Robin (RR) and Genetic Algorithm techniques. According to the simulation results, the suggested algorithm reduced overall costs by 13.38% and energy usage by 6.59% compared to the Genetic Algorithm (GA). Compared to RR, the proposed method minimizes energy use by 13.76% and total costs by 18.46%.

Author 1: Praveen Kumar Mishra

Author 2: Amit Kumar Chaturvedi

Keywords: IoT; fog computing; task scheduling; multi objective Model; iFogSim tool

Paper 44: Deep Learning with IoT-Based Solar Energy System for Future Smart Agriculture System

Abstract: Agriculture has a considerable contribution to the economy. Agriculture automation is a serious issue that is becoming more prevalent around the world. Farmers' traditional practices were insufficient to achieve these objectives. Artificial Intelligence (A1) and the Internet of Things (IoTs) are being used in agriculture to improve crop yield and quality. Distributed solar energy resources can now be remotely operated, monitored, and controlled through the IoT and deep learning technology. The development of an IoT-based solar energy system for intelligent irrigation is critical for water- and energy-stressed areas around the world. The qualitative design focuses on secondary data collection techniques. The deep learning model Radial Basis Function Networks (RBFN) is used in conjunction with the Elephant Search Algorithm (ESA) in this IoT-based solar energy system for future smart agriculture. Sensor systems help farmers understand their crops better, reduce their environmental impact and conserve resources. These advanced systems enable effective soil and weather monitoring, as well as water management. To provide the required operating power, the proposed system, RBFN-ESA, employs an IoT-based solar cell forecasting process. The proposed model RBFN-ESA will collect these data to predict the required parameter values for solar energy systems in future smart agriculture systems. The results of the RBFN-ESA model are effective and efficient. According to the findings, RBFN-ESA outperforms CNN, ANN, SVM, RF, and LSTM in terms of energy consumption (56.764J for 100 data points from the dataset), accuracy achieved (97.467% for 600 nodes), and soil moisture level (94.41% for 600 data).

Author 1: Vidya M S

Author 2: Ravi Kumar B. N

Author 3: Anil G. N

Author 4: Ambika G. N

Keywords: Precision agriculture; smart monitoring; Internet of Things; Radial Basis Function Networks; Elephant Search Algorithm (ESA)

Paper 45: Subjectivity Analysis of an Enhanced Feature Set for Code-Switching Text

Abstract: The phenomenon of code-switching has posed a new challenge to the linguistic computing area. Conventionally, the computer will process monolingual text or multilingual text. However, code-switching is different from this kind of text. Two or more languages are used to construct a piece of code-switching text, particularly a code-switching sentence. It is challenging for the computer to process a piece of code-switching text with languages that exist simultaneously. The challenge is more intense for the computer in subjectivity analysis, where the computer should distinguish subjective from objective code-switching text. This paper proposed three feature sets for subjectivity analysis on Malay-English code-switching text: Embedded Code-Switching Feature Sets, Unified Code-Switching Feature Sets, and Stylistic Feature Sets. These feature sets were enhanced from the monolingual feature set of subjectivity analysis. Experiments were conducted using the data harvested from Malay-English blogs. These data were labelled as either subjective or objective. Two machine learning classifiers – the Support Vector Machine (SVM) and Naive-Bayes, were used to evaluate the classification performance of the proposed feature sets. The experiments were carried out on individual feature sets and the combination of them. The results show the classification performance from combining the unified and stylistic feature sets surpassed other proposed feature sets at 59% accuracy. Therefore, it is concluded that the combination of unified and stylistic feature sets is necessary for the subjectivity analysis of Malay-English code-switching text.

Author 1: Emaliana Kasmuri

Author 2: Halizah Basiron

Keywords: Subjectivity analysis; code-switching; enhanced feature sets; Malay-English text

Paper 46: A Lightweight Privacy Preservation Protocol for IOT

Abstract: Due to rapid evolution of Internet of things (IOT) in terms of hardware, software and communication leads to widespread expansion across many domains and sectors. This expansion consequently results in sensitive data transfer increase for purposes of complex calculations and decision making which in turn leads to increase of data attacks and leakage which results in data privacy violation. Although, a lot of current solutions tried to fulfill data privacy via lightweight mechanisms but neither provided end to end protection nor gave a focus to metadata protection which can reveal valuable information about data it describes. This paper presents a lightweight complete data privacy protocol which manages the lifecycle of data starting from object registration till data transfer to cloud. The proposed protocol is a trusted third party free (TTP-Free) which adopts anonymization techniques, lightweight key agreement protocol, end to end encryption and message authentication code to fulfill identity and data protection which in turn fulfill complete data privacy.

Author 1: Ahmed Mahmoud Al-Badawy

Author 2: Mohammed Belal

Author 3: Hala Abbas

Keywords: IOT; data privacy; lightweight protocols; end to end protection; data and metadata protection

Paper 47: Basira: An Intelligent Mobile Application for Real-Time Comprehensive Assistance for Visually Impaired Navigation

Abstract: Individuals with visual impairments face numerous challenges in their daily lives, with navigating streets and public spaces being particularly daunting. The inability to identify safe crossing locations and assess the feasibility of crossing significantly restricts their mobility and independence. The profound impact of visual impairments on daily activities underscores the urgent need for solutions to improve mobility and enhance safety. This study aims to address this pressing issue by leveraging computer vision and deep learning techniques to enhance object detection capabilities. The Basira mobile application was developed using the Flutter platform and integrated with a detection model. The application features voice command functionality to guide users during navigation and assist in identifying daily items. It can recognize a wide range of obstacles and objects in real-time, enabling users to make informed decisions while navigating. Initial testing of the application has shown promising results, with clear improvements in users' ability to navigate safely and confidently in various environments. Basira enhances independence and contributes to improving the quality of life for individuals with visual impairments. This study represents a significant step towards developing innovative technological solutions aimed at enabling all individuals to navigate freely and safely.

Author 1: Amal Alshahrani

Author 2: Areej Alqurashi

Author 3: Nuha Imam

Author 4: Amjad Alghamdi

Author 5: Raghad Alzahrani

Keywords: Visual impairment; mobility application; computer vision; object detection; obstacle detection

Paper 48: Machine Translation-Based Language Modeling Enables Multi-Scenario Applications of English Language

Abstract: Traditional machine translation models suffer from problems such as long training time and insufficient adaptability when dealing with multiple English language scenarios. At the same time, some models often struggle to meet practical translation needs in complex language environments. A translation model that combines the feed-forward neural network decoder and the attention mechanism is suggested as a solution to this problem. Additionally, the model analyzes the similarity of the English language to enhance its translation ability. The resulting machine translation model can be applied to different English scenarios. The study's findings showed that the model performs better when the convolutional and attention layers have a higher number of layers relative to one another. The highest average value of the bilingual evaluation study for the research use model was 29.65. The research use model can machine translate different English language application scenarios and also the model performed better. The new model performed better than the traditional model and was able to translate the English language well in a variety of settings. The model used in the study had the maximum parameter data size of 4586, which is 932 higher than the lowest statistical machine translation model of 3654. The metric value was 3.96 higher than the statistical machine translation model. It is evident that investigating the use of the model can enhance the English language scene translation effect, with each scene doing well in translation. This provides new ideas for the direction of multi-scene application of machine translation language model afterwards.

Author 1: Shengming Liu

Keywords: Machine translation; decoder; English language; multi-scene; attention mechanism

Paper 49: Detecting Malware of Windows OS Using AI Classification for Image of Extracted Behavior Features

Abstract: Malware detection is crucial for protecting digital environments. Traditional methods involve static and dynamic analysis, but recent advancements leverage artificial intelligence (AI) to enhance detection accuracy. This study aims to improve malware detection by integrating dynamic malware analysis with AI-driven techniques. The primary challenge addressed is accurately classifying and detecting malware based on behavior extracted from isolated virtual machines. By analyzing 50 malware samples and 11 benign programs, we extract ten behavioral features such as process ID, CPU usage, and network connections. We employ text-based classification using feedforward neural networks (FNN) and recurrent neural networks (RNN), achieving accuracy rates of 56% and 68%, respectively. Additionally, we convert the extracted features into grayscale images for image-based classification with a convolutional neural network (CNN), resulting in a higher accuracy of 70.1%. This multi-modal approach, combining behavioral analysis with AI, not only enhances detection accuracy but also provides a comprehensive understanding of malware behavior compared to competing methods.

Author 1: Kang Dongshik

Author 2: Noor Aldeen Alhamedi

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

Paper 50: Dynamic Path Planning for Autonomous Robots in Forest Fire Scenarios Using Hybrid Deep Reinforcement Learning and Particle Swarm Optimization

Abstract: The growing frequency of forest area fires poses critical challenges for emergency response, necessitating progressive solutions for effective navigation and direction planning in dynamic environments. This study investigates an adaptive technique to enhance the performance of autonomous robots deployed in forest area fireplace scenarios. The primary objective is to develop a hybrid methodology that integrates advanced studying strategies with optimization techniques to enhance route planning beneath unexpectedly changing situations. To reap this, a simulation-based total framework became hooked up, in which self-reliant robots were tasked with navigating diverse forest fire eventualities. The method includes schooling a model to dynamically adapt to environmental modifications at the same time as optimizing direction choice in real time. Performance metrics together with direction efficiency, adaptability to obstacles, and reaction time been analyzed to assess the effectiveness of the proposed solution. Results indicate an enormous improvement in path planning performance as compared to traditional methods, with more suitable adaptability main to faster response instances and extra effective navigation. The findings underscore the functionality of the proposed method to cope with the complexities of forest area fire environments, demonstrating its potential for real-world applications in disaster response. The results are shown in the conceived DRL-PSO framework where execution time is reduced up to 95% and the success rate of 95 % for the proposed method compared to the conventional ones. Python is used to implement the proposed work. Compared to the proposed method’s execution time of 68. 3 seconds and the highest success rate among evaluated strategies, so it can be used as a powerful solution for autonomous drone navigation in dangerous situations. In the end, this research contributes precious insights into adaptive route planning for self-sufficient robots in unsafe situations, providing a strong framework for destiny advancements in disaster management technologies.

Author 1: N. K. Thakre

Author 2: Divya Nimma

Author 3: Anil V Turukmane

Author 4: Akhilesh Kumar Singh

Author 5: Divya Rohatgi

Author 6: Balakrishna Bangaru

Keywords: Adaptive path planning; deep reinforcement learning; disaster environments; drone rescuing; particle swarm optimization; forest fire

Paper 51: A Convolutional Neural Network-Based Predictive Model for Assessing the Learning Effectiveness of Online Courses Among College Students

Abstract: With the development of artificial intelligence (AI) technology, higher education institutions usually consider both online courses and offline classrooms in the course design process. To verify the effectiveness of online courses, this study designed a deep learning model to analyze the learning behavior of online course users (college students) and predict their final grades. Firstly, our method summarizes several learning features that are used in machine learning models for predicting student grades, including the performance of users (college students) in online courses and their basic information. Based on nutcracker optimization algorithm (NOA), we designed a multi-layer convolutional neural network (CNN) and developed an improved NOA (I-NOA) to optimize the internal parameters of the CNN. Prediction mainly includes two steps: firstly, analyzing users' emotions based on their comments in online course forums. Secondly, predict the final grade based on the user's emotions and other quantifiable learning features. To validate the effectiveness of INOA-Based CNN (I-NOA-CNN) algorithm, we evaluated it using a dataset consisting of five different online courses and a total of 120 students. The simulation results indicate that compared with existing methods, the I-NOA-CNN algorithm has higher prediction accuracy, and the proposed model can effectively predict the learning effect of users.

Author 1: Xuehui Zhang

Author 2: Lin Yang

Keywords: Convolutional neural network; nutcracker optimization algorithm; assessment of learning effectiveness; college students; online courses

Paper 52: Forensic Facial Reconstruction from Sketch in Crime Investigation

Abstract: Many crimes are committed every day all over the world, and one of them is a criminal offense that includes a wide range of illegal acts such as murder, theft, assault, rape, kidnapping, fraud, and others. Criminals pose a threat to security, which harms the public interest. In this case, the police question all eyewitnesses at the crime scene, and sometimes, witnesses who were present at the crime scene can remember the face of the criminal. The witness accurately describes the person's facial features in the report, such as eyes, nose, etc. Law enforcement authorities use eyewitness information to identify the person. Criminal investigations can be accelerated by converting sketched faces into actual images, but this requires eyewitnesses to confirm the description in the report. Drawings make it very difficult to identify real human faces because they do not contain the details that help to catch criminals. In contrast, color photographs contain many details that help to identify facial features more clearly. This work proposes to generate color images using the modified modulation Sketch-to-Face CycleGAN and then pass them through Generative Facial Prior-GAN. CycleGAN consists of a generator and discriminator. The generator is used to generate colored images, and the discriminator is used to identify whether the images are real or fake. These are then passed to GFPGAN to improve the quality of the colored images. The structural similarity index measure of 0.8154 is achieved when creating photorealistic images from drawings.

Author 1: Doaa M. Mohammed

Author 2: Mostafa Elgendy

Author 3: Mohamed Taha

Keywords: Sketch-to-Face; facial features; Sketch-to-Face CycleGAN; victim's identification; criminal offense

Paper 53: Artificial Intelligence-Driven Decision Support Systems for Sustainable Energy Management in Smart Cities

Abstract: Due to the ongoing urbanization trend, smart cities are critical to designing a sustainable future. Urban sustainability involves action-oriented approaches for optimizing resource usage, ecological impact reduction, and overall efficiency enhancement. Energy management is one of the main concerns in urban, residential, and building planning. Artificial Intelligence (AI) uses data analytics and machine learning to instigate business automation and deal with intelligent tasks involved in numerous industries. Thus, AI needs to be considered in the strategic plan, especially in the long-term strategy of smart city planning. Decision Support Systems (DSS) are integrated with human-machine interaction methods like the Internet of Things (IoT). Along with their growth in size and complexity, the communications of IoT smart devices, industrial equipment, sensors, and mobile applications present an increasing challenge in meeting Service Level Agreements (SLAs) in diverse cloud data centers and user requests. This challenge would be further compounded if the energy consumption of industrial IoT networks also increased tremendously. Thus, DSS models are necessary for automated decision-making in crucial IoT settings like intelligent industrial systems and smart cities. The present study examines how AI can be integrated into DSS to tackle the intricate difficulties of sustainable energy management in smart cities. The study examines the evolution of DSSs and elucidates how AI enhances their functionalities. The study explores several AI methods, such as machine learning algorithms and predictive analytics, that aid in predicting, optimizing, and making real-time decisions inside urban energy systems. Furthermore, real-world instances from different smart cities highlight the practical applications, benefits, and interdisciplinary collaboration necessary to successfully implement AI-driven DSS in sustainable energy management.

Author 1: Ning MA

Keywords: Smart cities; artificial intelligence; decision support systems; sustainable energy management; urban resilience; interdisciplinary collaboration

Paper 54: Exploring Multimedia Movement Through Spatio-Temporal Indexing and Double-Cache Schemes

Abstract: Conventional IP/TCP designs encounter several safety and scalability concerns with the growing demand for application services. A novel Internet design, like a Content Center Network (CCN), was introduced to address these issues comprehensively. Every hub within a CCN is responsible for data storage. The collaboration guarantees users quick data retrieval. By collaborating with dual caches, network peers can access data from their caches and leverage other peers' caches, resulting in improved cache utilization and overall network speed. The present study examines multimodal digital artworks' form, style, and action relationships and views them as holistic creative units. The study examines the complex structure of digital content following current information. We present a distributed index incorporating spatio-temporal information to address the challenges of storing and retrieving large amounts of spatio-temporal data. This distributed index combines internal R with external B+ trees to provide high concurrency and low latency indexing services for external applications. With double buffer technology and distributed index architecture, we can optimize the cache utility of content center networks and enhance the retrieval speed of multimedia data. Adopting the distributed index, designed to accommodate spatio-temporal data in the multimedia digital art design, can enhance large-scale storage and retrieval for Internet-future architectures.

Author 1: Zhen QIN

Author 2: Lin ZHANG

Keywords: Multimedia digital art; double-cache collaboration; distributed indexing; spatio-temporal data; content center network

Paper 55: A Secure and Efficient Framework for Multi-User Encrypted Cloud Databases Supporting Single and Multiple Keyword Searches

Abstract: Multi-user encrypted cloud databases have become essential for secure data storage and retrieval, especially when supporting both single and multiple keyword searches. Ensuring data confidentiality, integrity, and efficient access within such systems is paramount, particularly when dealing with multiple data owners and users. This paper presents a Secure Encrypted Trie-based Search (SETBS) method that significantly enhances multi-owner authentication, data secrecy, and data integrity in cloud environments. The SETBS framework leverages a sophisticated Merkle hash tree for dynamic maintenance and autonomous user verification, ensuring that the identity of users is reliable and that personal information remains protected across various ownership domains. By optimally utilizing resources, SETBS provides a robust and efficient solution for managing data in cloud environments. The framework addresses the bottleneck issue by distributing the workload among first-level owners, resulting in fair resource distribution and increased system efficiency. A key feature of the SETBS method is its ability to guarantee data integrity without compromising security. Users can be assured that their data remains unaltered and protected from unauthorized access, thanks to the integration of the Merkle hash tree. This mechanism enables clients to confirm the integrity of their data stored in the cloud, providing peace of mind regarding its security. Moreover, SETBS proves to be a flexible and scalable solution for large-scale cloud deployments, efficiently managing multiple data owners and parallelizing the processing load. The framework's focus on data privacy ensures that personal data remains secure during search operations. With lower encryption and decryption times compared to existing methods such as SPEKS, DSSE, and MKHE, SETBS demonstrates superior performance and is implemented in Python. This comprehensive approach offers an all-encompassing solution for businesses seeking to enhance their cloud security architecture while ensuring efficient data management, from processing to real-time or batch data analysis.

Author 1: J V S Arundathi

Author 2: K V V Satyanarayana

Keywords: Secure keyword search; encrypted search; multi-user framework; encrypted cloud database; single and multiple key users

Paper 56: Exploring the Application of Neural Networks in the Learning and Optimization of Sports Skills Training

Abstract: Sports skills training is a crucial component of sports education, significantly contributing to the development of athletic abilities and overall physical literacy. It is essential to utilized neural networks to optimize traditional training methods that are inefficient and rely on subjective assessments. This paper develops methods for sports action recognition and athlete pose estimation and prediction based on deep neural networks. Given the complexity and rapid changes in sports skills, we propose a multi-task framework-based HICNN-PSTA model for jointly recognizing sports actions and estimating human poses. This method leverages the advantages of Convolution and Involution operators in computing channel and spatial information to extract sports skill features and uses a decoupled multi-head attention mechanism to fully capture spatio-temporal information. Furthermore, to accurately predict human poses to avoid potential sports injuries, this paper introduces an MS-GCN prediction model based on the multi-scale graph. This method utilizes the constraints between human body key points and parts, dividing the 2D human pose into different levels, significantly enhancing the modeling capability of human pose sequences. The proposed algorithms have been thoroughly validated on a basketball skills dataset and compared with various advanced algorithms. Experimental results sufficiently demonstrate the effectiveness of the proposed methods in sports action recognition and human pose estimation and prediction. This research advances the application of deep neural networks in the field of sports training, providing significant reference value for related studies.

Author 1: Dazheng Liu

Keywords: Deep neural network; action recognition; 2D pose prediction; pose estimation; sports skill training; attention mechanism

Paper 57: Birth Certificates Delivery, Traceability and Authentication Using Blockchain Technology

Abstract: Now-a-days, the vast majority of birth certificate registration systems are paper-based and managed independently by administrative communities. This means that birth information only exists at the place where the birth is registered, which facilitates the counterfeiting or falsification of such identity documents. Therefore, the implementation of a system for the issuance, traceability, and authentication of birth certificates is imperative. Blockchain, characterized by transparency, immutability, protection, privacy, and autonomy, makes this technology the ideal solution for implementing a birth certificate registration, traceability, and authentication system. This article presents a decentralized system for the registration, traceability, and authentication of birth certificates based on Hyperledger Fabric private blockchain deployed in a Virtual Private Network - Multi-Protocol Label Switching (VPN-MPLS) network. This birth certificate is characterized on one hand by the attributes of its owner and on the other hand by a Quick Response (QR) code containing the digital signature of its signer and the unique identifier of the birth certificate. Within the network, the unique identifier of the generated document is hashed and stored using the Secure Hash Algorithm-256 (SHA-256) hash function to optimize storage space and enhance security. Furthermore, the proposed platform includes an application designed using Docker Compose, Apache CouchDB, NodeJS, Go, and Hyperledger Explorer. The designed model is a birth certificate registration platform that ensures enhanced security and transparency.

Author 1: Tankou Tsomo Maurice Eddy

Author 2: Bell Bitjoka Georges

Author 3: Ngohe Ekam Paul Salomon

Author 4: Ekani Mebenga Vianney Boniface

Keywords: Birth certificates; blockchain; security; traceability; authentication; counterfeiting; falsification; hyperledger fabric

Paper 58: Optimizing Customer Interactions: A BERT and Reinforcement Learning Hybrid Approach to Chatbot Development

Abstract: In the case of chatbots massive progress has been made, but problems remain in handling the complexity of the sentence and in context relevance. Traditional models can be rather insufficient when it comes to providing various levels of detail in the responses to the end-users’ questions, particularly when referring to customer support scenarios. To overcome these limitations, this research comes up with a new model which combines the BERT model with DRL. Through DRL, BERT pre-training is adding flexibility and correspondence to correctly perceive contextual delicate matters in the response. The proposed method includes the following pipeline where in; data tokenization, conversion to lowercase characters, lemmatization and then passes through the BERT fine-tuned model. DRL is utilized to optimize the chatbot’s response in the light of long term rewards and the conversational history, the interactions are formulated as a Markov Decision Process with the reward functions based on cosine similarity of the consecutive responses. This makes it feasible for the chatbot to provide context based replies in addition to the option of constant learning for enhanced performance. It also proved that the accuracy and relevance of the BERT-DRL hybrid system were higher than traditional models according to the BLEU and ROUGE scores. The performance of the chatbot also increases with the length of the conversation and the transitions from one response to the other are coherent. This research contributes to the field through the integration of BERT in understanding language and DRL in the iterative learning process in the innovation within the flaws of chatbot technologies and establishing a new benchmark for conversational AI in customer service settings.

Author 1: K. R. Praneeth

Author 2: Taranpreet Singh Ruprah

Author 3: J Naga Madhuri

Author 4: A L Sreenivasulu

Author 5: Syed Shareefunnisa

Author 6: Vuda Sreenivasa Rao

Keywords: Chatbots; BERT (Bidirectional Encoder Representations from Transformers); RL (Reinforcement Learning); customer service; responsiveness

Paper 59: Liver and Tumour Segmentation Using Anchor Free Mechanism-Based Mask Region Convolutional Neural Network

Abstract: An accurate liver tumour segmentation helps acquire the measurable biomarkers for decision support systems and Computer-Aided Diagnosis (CAD). However, most existing approaches fail to effectively segment tumours in the liver due to the overlapping of liver with any other organ in the image. To solve this problem, this research proposes Anchor Free with Masked Region-based Convolutional Neural Network (AFMRCNN) approach for segmenting liver tumours. The AF attains a precise localization of tumours by directly predicting the tumour location without relying on predefined anchor boxes. Standard datasets like LiTS and CHAOS are utilized to experiment with the efficiency of the proposed method. An EfficientNetB2 is performed to extract the most relevant features from the segmented data. The Deep Neural Network (DNN) is performed for the classification of liver tumours into binary classes by capturing intricate patterns and relationships in the data with the help of a non-linear activation function. The experimental results exhibit the proposed ARMRCNN method’s commendable segmentation performance of 0.998 Dice Similarity Coefficient (DSC), as opposed to the existing methods, UoloNet and UNet++ + pre-activated multiscale Res2Net approach with Channel-wise Attention (PARCA) on the LiTS dataset.

Author 1: Sangi Narasimhulu

Author 2: Ch D V Subba Rao

Keywords: Anchor free; computer-aided diagnosis; deep neural network; EfficientNetB2; liver and tumor segmentation; masked region-based convolutional neural network

Paper 60: Efficient Task Offloading Using Ant Colony Optimization and Reptile Search Algorithms in Edge Computing for Things Context

Abstract: The widespread use of Internet of Things (IoT) technology has triggered unparalleled data creation and processing needs, necessitating effective computation offloading solutions. Conventional edge computing approaches have difficulties in dealing with rising energy usage issues and task allocation delays. This study introduces a novel hybrid metaheuristic algorithm called ACO-RSA, which synergizes two metaheuristic algorithms, Ant Colony Optimization (ACO) and Reptile Search Algorithm (RSA). The proposed approach addresses the energy and latency issues associated with offloading computations in IoT edge computing environments. A comprehensive system design that effectively encapsulates the uplink transmission communication model and a personalized multi-user computing task load model is developed. The system considers various constraints, such as network latency, task complexity, and available computing resources. Based on this, we formulate an optimization objective suitable for computing outsourcing in the IoT ecosystem. Simulations conducted in a real-world IoT scenario demonstrate that ACO-RSA significantly reduces both time delay and energy consumption compared to benchmark algorithms, achieving up to 27.6% energy savings and 25.4% reduction in time delay. ACO-RSA exhibits robustness and scalability when optimizing task offloading in IoT edge computing environments.

Author 1: Ting Zhang

Author 2: Xiaojie Guo

Keywords: Task offloading; edge computing; ant colony optimization; reptile search algorithm; Internet of Things; energy efficiency

Paper 61: Bubble Detection in Glass Manufacturing Images Using Generative Adversarial Networks, Filters and Channel Fusion

Abstract: With the increasing production of glassware products, the detection of bubble defects has been of vital importance. The manual inspection of glass bubble defects is considered to be tedious and inefficient way due to the increasing volume of images, and the high probability of human error. Computer vision-based methods provide us with a platform for automating the bubble defect detection process which can overcome the disadvantages associated with manual inspection thereby significantly reducing the cost and improving the quality. To address these issues, we propose an integrated deep learning (DL) based bubble detection algorithm, in which an image data set is prepared using a Generative Adversarial Network (GAN). The proposed algorithm exploits the Information-Preserving Feature Aggregation (IPFA) module for achieving semantic feature extraction by maintaining the small defects’ internal features. To weed out irrelevant information due to fusion, the proposed research introduces the Conflict Information Suppression Feature Fusion Module (CSFM) to further advance the component combination methodology, the Fine-Grained Conglomeration Module (FGAM) is employed to facilitate cooperation among feature maps at various levels. This approach mitigates the generation of conflicting information arising from erroneous features. The algorithm improved performance with an accuracy rate of 0.677 and a recall rate of 0.716 with a precision value of 0.638.

Author 1: Md Ezaz Ahmed

Author 2: Mohammad Khalid Imam Rahmani

Author 3: Surbhi Bhatia Khan

Keywords: Computer vision; Generative Adversarial Network; Information-Preserving Feature Aggregation; Conflict Information Suppression Feature Fusion Module; Fine-Grained Aggregation Module; deep learning

Paper 62: Dimensionality Reduction Evolutionary Framework for Solving High-Dimensional Expensive Problems

Abstract: Most of improvement strategies for surrogate-assisted optimiza-tion algorithms fail to help the population quickly locate satis-factory solutions. To address this challenge, a novel framework called dimensionality reduction surrogate-assisted evolutionary (DRSAE) framework is proposed. DRSAE introduces an effi-cient dimensionality reduction network to create a low-dimensional search space, allowing some individuals to search in the population within the reduced space. This strategy signifi-cantly lowers the complexity of the search space and makes it easier to locate promising regions. Meanwhile, a hierarchical search is conducted in the high-dimensional space. Lower-level particles indiscriminately learn from higher-level peers, corre-spondingly the highest-level particles undergo self-mutation. A comprehensive comparison between DRSAE and mainstream HEPs algorithms was conducted using seven widely used benchmark functions. Comparison experiments on problems with dimensionality increasing from 50 to 200 further substanti-ate the good scalability of the developed optimizer.

Author 1: SONGWei

Author 2: ZOUFucai

Keywords: Dimensionality reduction; high-dimensional expensive optimiza-tion; Surrogate-assisted model

Paper 63: Optimization of the Energy-Saving Data Storage Algorithm for Differentiated Cloud Computing Tasks

Abstract: This study presents a novel energy-saving data storage algorithm designed to enhance data storage efficiency and reduce energy consumption in cloud computing environments. By intelligently discerning and categorizing various cloud computing tasks, the algorithm dynamically adapts data storage strategies, resulting in a targeted optimization methodology that is both devised and experimentally validated. The study findings demonstrate that the optimized model surpasses comparative models in accuracy, precision, recall, and F1-score, achieving peak values of 0.863, 0.812, 0.784, and 0.798, respectively, thereby affirming the efficacy of the optimized approach. In simulation experiments involving tasks with varying data volumes, the optimized model consistently exhibits lower latency compared to Attention-based Long Short-Term Memory Encoder-Decoder Network and Deep Reinforcement Learning Task Scheduling models. Furthermore, across tasks with differing data volumes, the optimized model maintains high throughput levels, with only marginal reductions in throughput as data volume increases, indicating sustained and stable performance. Consequently, this study is pertinent to cloud computing data storage and energy-saving optimization, offering valuable insights for future research and practical applications.

Author 1: Peichen Zhao

Keywords: Energy-saving data storage algorithm; differentiated task recognition; cloud computing; intelligent storage strategy; data classification and distribution

Paper 64: Advancing Quantum Cryptography Algorithms for Secure Data Storage and Processing in Cloud Computing: Enhancing Robustness Against Emerging Cyber Threats

Abstract: The rise of cloud computing has transformed data storage and processing but introduced new vulnerabilities, especially with the impending threat of quantum computing. Traditional cryptographic methods, though currently effective, are at risk of being compromised by quantum attacks. This research aims to develop a quantum-resistant security framework for cloud environments, combining lattice-based cryptography with Quantum Key Distribution (QKD) protocols, particularly the E91 protocol, for secure key management. The framework also incorporates quantum authentication protocols to enhance user identity verification, protecting against unauthorized access and tampering. The proposed solution balances robust security with practical implementation, ensuring scalability and efficiency in real-world cloud environments. Performance evaluations indicate an encryption time of approximately 30 milliseconds, outperforming existing methods such as RSA and DES. This research contributes to the development of future-proof cryptographic standards, addressing both current security challenges and emerging quantum computing threats. By leveraging quantum mechanics, the framework strengthens cloud-based data protection, providing a resilient solution against evolving cyber risks. The results hold significant promise for advancing cloud security, laying the groundwork for next-generation encryption techniques that can withstand the threats posed by quantum computing.

Author 1: Devulapally Swetha

Author 2: Shaik Khaja Mohiddin

Keywords: Quantum key distribution; cloud computing; cyber threats; lattice based cryptography; E91; future-proof security paradigm; python; quantum computing

Paper 65: Advancements and Challenges in Geospatial Artificial Intelligence, Evaluating Support Vector Machines Models for Dengue Fever Prediction: A Structured Literature Review

Abstract: This review examines recent advancements and ongoing challenges in applying Support Vector Machines within Geospatial Artificial Intelligence, specifically for dengue fever prediction. Recent developments in Support Vector Machines include the introduction of advanced kernel methods, such as Radial Basis Function and polynomial kernels, which enhance the model’s ability to handle complex spatial data and interactions. Integration with high-resolution geospatial data and real-time analytics has significantly improved predictive accuracy, particularly in mapping environmental factors influencing disease spread. However, challenges persist, including issues with data quality, computational demands, and model interpretability. Data scarcity and the high computational cost of Support Vector Machines, especially with non-linear kernels, necessitate optimization techniques and advanced computing resources. Parameter tuning and enhancing model interpretability are critical for effective implementation. Future research should focus on developing new kernels and hybrid models that combine Support Vector Machines with other machine learning approaches to address these challenges. Practical applications in public health can benefit from improved real-time data processing and high-resolution analytics, while ensuring adherence to ethical and regulatory standards. This review underscores the potential of Support Vector Machines in Geospatial Artificial Intelligence for disease prediction and highlights areas where further innovation and research are needed to enhance its practical utility in public health.

Author 1: Hetty Meileni

Author 2: Ermatita

Author 3: Abdiansah

Author 4: Nyayu Latifah Husni

Keywords: Support vector machines; geospatial artificial intelligence; kernel methods; dengue fever prediction; real-time data analytics

Paper 66: Multimedia Network Data Fusion System Integrating SSA and Reinforcement Learning

Abstract: To improve the performance and efficiency of multimedia network data fusion system, this study proposes an improved sparrow search algorithm on the ground of reinforcement learning algorithm and sparrow search algorithm, and improves the multimedia network data fusion model on the ground of this algorithm. A performance comparison experiment was conducted on the improved sparrow search algorithm, and it was found that the algorithm entered a convergence state after 380 iterations in a unimodal function. Its time consumption is lower than other comparison algorithms, and it has not fallen into the local optimal situation after 500 iterations in the multimodal benchmark function. Its performance is significantly superior to other comparison algorithms. Moreover, the study conducted relevant experiments on the multimedia network data fusion model and found that the F1 value output by the model was 0.37, with an accuracy of 92.4%, which is higher than other data fusion models. And the mean square error of this model reaches 0.52, and the processing time is 0.1 seconds, which is lower than other comparative data fusion models. The quality of output data and data processing efficiency of this model are better. The relevant outcomes demonstrate that the improved sparrow search algorithm possesses good global search and convergence performance. And the improved multimedia network data fusion model has better accuracy and efficiency, and has good practical application value. This study can provide reference and reference for multimedia network data fusion systems.

Author 1: Fangrui Li

Keywords: Sparrow search algorithm; reinforcement learning; multimedia network; data fusion; performance improvement

Paper 67: Application of U-Net Network Algorithm in Electronic Information Field

Abstract: This rapidly evolving landscape, which includes the field of medical diagnostics, has integrated with the electronic data (E-Data) field to provide precise and efficient treatment for complex medical conditions. The research field has further catapulted its reach to include various data types, including image, video, medical expert diagnostic type, and sensor input, out of which the image-based diagnostic model has excellent research potential. Convolutional Neural Network (CNN) based models have evolved into better Deep Learning (DL) models for handling complex intricacies featured in the input image. U-Net is a prominent CNN model developed to handle the features of image data. The U-Net excels in capturing detailed features through its encoder-decoder structure and skip connections, but its uniform weighting across different network layers may not adequately address the subtleties involved in complex medical anomaly detection. This work proposed the Attention Calibrated U-Net (ACU-Net) model that is designed to address the challenges of U-Net in detecting Fetal Cardiac Rhabdomyoma (FCR) from echocardiographic (ECG) images. FCR is a prevalent benign cardiac tumor in fetuses that poses significant diagnostic challenges due to its variable manifestations and the intricate nature of fetal cardiac anatomy. The proposed model enhances the U-Net architecture with attention mechanisms and employs a hybrid Loss Function (LF) that combines Cross-Entropy Loss, Dice Loss, and an attention-driven component for effective FCR detection. The model was compared against others and demonstrated better specificity, accuracy, precision, recall, and F1-score performance across various ECG views (LVOT, RVOT, 3VT, and 4CH).

Author 1: Liang Wang

Keywords: U-Net; attention calibrated U-Net; convolutional neural network; deep learning; digital data; accuracy

Paper 68: Natural Disaster Clustering Using K-Means, DBSCAN, SOM, GMM, and Mean Shift: An Analysis of Fema Disaster Statistics

Abstract: Natural disasters tend to ruin people’s lives and infrastructure, which requires comprehensive analysis and understanding to inform effective disaster management and response planning. This research addresses the lack of in-depth analysis of federally declared disasters in the United States using a dataset sourced from FEMA. Through the application of unsupervised learning techniques, including K-means clustering, DBSCAN, self-organizing maps (SOM), and the Gaussian mixture model (GMM), similar types of disasters are clustered based on their frequency. The relationship between disaster type and disaster frequency is analyzed to gain insight into patterns and correlations, facilitating targeted mitigation and adaptation strategies. By using the techniques of clustering, we can accurately group similar disaster types, duration time, occurring time and location of disaster. By implementing these approaches, our study aims to improve the understanding of disaster occurrences and inform decision-making processes in disaster mitigation strategies and adaptation strategies.

Author 1: Ting Tin Tin

Author 2: Yap Jia Hao

Author 3: Yong Chang Yeou

Author 4: Lim Siew Mooi

Author 5: Goh Ting Yew

Author 6: Temitope Olumide Olugbade

Author 7: Ali Aitizaz

Keywords: Natural disasters; disaster management; unsupervised learning; clustering; disaster frequency; disaster types; mitigation strategies; adaptation strategies

Paper 69: Impact of Read Theory in Mobile-Assisted Language Learning on Engineering Freshmen's Reading Comprehension Using BI-LSTM

Abstract: The effect of Read Theory in Mobile-Assisted Language Learning (MALL) on reading comprehension is critical, especially for engineering freshmen who require excellent language abilities to navigate their complicated academic courses. Read Theory is a customized reading platform that offers adaptive reading activities based on the user's ability level, which is especially useful in MALL settings where accessibility and flexibility are essential. However, traditional methods of MALL have frequently faced with constraints, such as the inability to completely adapt to students' different and dynamic learning demands. This deficiency usually results in poor improvement in reading skills because the conventional paradigms do not capture the intricate and diverse learning processes that are necessary for the effective learning of languages. To fix these issues, a Deep Learning approach that involves the implementation of BI-LSTM networks for enhancing the completion’s reading outcomes is offered. BI-LSTM is more suitable for this task because it has forward and backward reading capabilities to better understand and predict the dynamics of language acquisition. The research established improvement and an astonishing accuracy of 99.3%. The implementation was done using Python. This high value of accuracy disproves the common weakness of the strategy and provides convincing evidence that the proposed approach can significantly enhance MALL projects’ outcomes. The specified technique, which improves on the flaws of previous approaches, does not only improve the process of reading but has the potential to revolutionize language acquisition for engineering students, making it more effective and conforming to ability.

Author 1: E. Pearlin

Author 2: S. Mercy Gnana Gandhi

Keywords: Read theory; language learning; Bi-LSTM; mobile-assisted; deep learning

Paper 70: Advancing Natural Language Processing with a Combined Approach: Sentiment Analysis and Transformation Using Graph Convolutional LSTM

Abstract: Sentiment analysis is a key component of Natural Language Processing (NLP), taking into account the extraction of emotional cues from text. However, traditional strategies often fail to capture diffused feelings embedded in language. To deal with this, we advocate a novel hybrid model that complements sentiment analysis by way of combining Graph Convolutional Networks (GCNs) with Long Short-Term Memory (LSTM) networks. This fusion leverages LSTM’s sequential reminiscence abilities and GCN’s ability to model contextual relationships, allowing the detection of nuanced feelings regularly overlooked with the aid of conventional techniques. The hybrid technique demonstrates superior generalization overall performance and resilience, making it mainly powerful in complicated sentiment detection responsibilities that require a deeper knowledge of text. These results emphasize the capacity of combining sequential memory architectures with graph-based contextual facts to revolutionize sentiment analysis in NLP. This study not only introduces an innovative approach to sentiment analysis but also underscores the importance of integrating advanced techniques to push the boundaries of NLP research. This cutting-edge hybrid model surpasses the performance of previous techniques like CNN, CNN-LSTM, and RNN-LSTM with an amazing accuracy of 99.33%, creating a new benchmark in sentiment analysis. The results demonstrate how more precise sentiment analysis made possible by fusing sequential memory architectures with graph-based contextual information might revolutionise NLP. The findings provide a new benchmark, advancing the sphere by way of enabling greater specific and nuanced sentiment evaluation for a wide range of programs, inclusive of purchaser remarks analysis, social media monitoring, and emotional intelligence in AI structures.

Author 1: Kedala Karunasree

Author 2: P. Shailaja

Author 3: T Rajesh

Author 4: U. Sesadri

Author 5: Choudaraju Neelima

Author 6: Divya Nimma

Author 7: Malabika Adak

Keywords: Graph Convolutional Networks (GCN); Long Short-Term Memory (LSTM); Natural Language Processing (NLP); sentiment analysis; emotions; text classification; machine learning

Paper 71: ABC-Optimized CNN-GRU Algorithm for Improved Cervical Cancer Detection and Classification Using Multimodal Data

Abstract: Cervical cancer is the second most common malignancy among women, making it a major public health problem worldwide. Early detection of cervical cancer is important because it increases the chances of effective treatment and survival. Regular screening and early management can prevent the growth of cervical cancer, thus reducing mortality. Traditional methods of detection, such as Pap smears, have proven useful, but are time-consuming and rely on behavioral interpretation by cytologists. To overcome these issues the study uses method another for a convolutional neural networks (CNNs) and gated recurrent units (GRUs) to detect and classify cervical cancer in Pap smear images by tuning with Artificial Bee Colony (ABC) Optimizer. This study used several datasets with high-resolution images from the SipakMed collection, with 4049 images and a fetal dataset with patient information for the CNN component of the model, specifically the ResNet-152 system, is extracted spatial attributes from these images. After feature extraction, the GRU component analyzes the sequential data to identify temporal combinations and patterns. This hybrid CNN-GRU algorithm uses the features of two networks: the ability of CNN to learn spatial patterns and the ability of GRU to understand sequential networks and tuning the parameters using ABC. The proposed model outperformed the conventional ML methods with a classification accuracy of 94.89%, and provided a reliable solution for early detection of cervical cancer Using these DL methods role which, not only enables a more accurate diagnosis, but also allows a comprehensive examination of the abnormal cervical cells, making it a positive detections to programs and patient outcomes. This work highlights the promise of cutting-edge AI techniques to improve cervical cancer diagnosis, and the need for faster and more accurate diagnosis in the battle to emphasize the fight against this common disease.

Author 1: Donepudi Rohini

Author 2: M Kavitha

Keywords: Cervical cancer; CNN-GRU; Pap smear images; Artificial Bee Colony Optimizer; early detection

Paper 72: Enhancing Student Well-Being Prediction with an Innovative Attention-LSTM Model

Abstract: This study introduces a groundbreaking method for predicting scholar well-being with the use of a sophisticated interest-primarily based Long Short-Term Memory (LSTM) version. Addressing the developing problem of intellectual health in academic settings, the studies pursuits to provide new insights and powerful techniques for reinforcing pupil mental well-being. The recognition is on enhancing the prediction of mental fitness issues via the revolutionary use of interest-primarily based LSTM algorithms, which excel in discerning various ranges of relevance among input facts points. The version leverages a unique methodology to procedure various datasets, which include academic information, social media activity, and textual survey responses. By emphasizing sizable capabilities like language patterns and shifts in educational performance, the attention-based totally LSTM version overcomes barriers of conventional predictive techniques and demonstrates superior accuracy in figuring out subtle indicators of mental health troubles. The schooling dataset is categorized into behavioral states along with "healthy," "confused," "traumatic," and "depressed," allowing the version to build a strong learning foundation. This research highlights the transformative ability of superior interest-primarily based strategies, offering an effective device for improving our know-how and predictive capabilities concerning adolescent mental fitness situations. The study underscores the significance of integrating progressive device studying tactics in addressing intellectual health demanding situations and enhancing standard scholar well-being. Upon implementation and rigorous checking out in Python, the proposed technique achieves a notable accuracy price of 98.9% in identifying mental fitness issues among college students. This observe underscores the transformative potential of superior interest-based totally strategies, thereby improving the expertise and predictive competencies concerning mental fitness conditions in teens.

Author 1: Vinod Waiker

Author 2: Janjhyam Venkata Naga Ramesh

Author 3: Ajmeera Kiran

Author 4: Pradeep Jangir

Author 5: Ritwik Haldar

Author 6: Padamata Ramesh Babu

Author 7: E. Thenmozhi

Keywords: Student mental health; attention-based LSTM; well-being enhancement; predictive modelling; innovative techniques

Paper 73: A Hybrid Intelligent System for IP Traffic Classification

Abstract: The classification of IP traffic is important for many reasons, including network management and security, quality of service (QoS) monitoring and provisioning, and high hardware utilisation. Recently, many machine learning-based IP traffic classifiers have been developed. Unfortunately, most of them need to be trained on large datasets and thus require a long training time and significant computational power. In this paper, I investigate this problem and, as a solution, present a hybrid system, which I call the ISITC, that combines the random forest (RF) and XGBoost (XGB) machine learning techniques with the support vector classifier (SVC) as the final estimator, the stacking classifier. This design leads to the development of a model that performs the classification of IP traffic and internet applications efficiently and with high accuracy. I evaluate the performance of the ISITC and various IP traffic classifiers, including neural network (NN), RF, decision tree (DT), and XGB classifiers and SVCs. The experimental results show that the ISITC provides the best IP traffic classification, with an accuracy of 96.7, and outperforms the other IP traffic classifiers: the NN classifier has an accuracy of 59, the RF classifier has an accuracy of 88.5, the DT classifier has an accuracy of 90.5, the XGB classifier has an accuracy of 89.8, and the SVC has an accuracy of 64.8.

Author 1: Muhana Magboul Ali Muslam

Keywords: Internet application classification; IP traffic classification; machine learning; machine learning techniques; stacking classifier

Paper 74: Deep Learning-Driven Localization of Coronary Artery Stenosis Using Combined Electrocardiograms (ECGs) and Photoplethysmograph (PPG) Signal Analysis

Abstract: The application of artificial intelligence (AI) to electrocardiograms (ECGs) and photoplethysmograph (PPG) for diagnosing significant coronary artery disease (CAD) is not well established. This study aimed to determine whether the combination of ECG and PPG signals could accurately identify the location of blocked coronary arteries in CAD patients. Simultaneous measurement of ECG and PPG signal data were collected from a Malaysian university hospital, including patients with confirmed significant CAD based on invasive coronary angiography. ECG and PPG datasets were concatenated to form a single dataset, thereby enhancing the information available for the training process. Experimental results demonstrate that the Convolutional Neural Networks (CNN) + Long Short-Term Memory (LSTM) + Attention (ATTN) mechanisms model significantly outperforms standalone CNN and CNN + LSTM models, achieving an accuracy of 98.12% and perfect Area Under the Curve (AUC) scores of 1.00 for the detection of blockages in the left anterior descending (LAD) artery, left circumflex (LCX) artery, and right coronary artery (RCA). The integration of LSTM layers captures temporal dependencies in the sequential data, while the attention mechanism selectively highlights the most relevant signal features. This study demonstrates that AI-enhanced models can effectively analyze simultaneous measurement of standard single-lead ECGs and PPG to predict the location of coronary artery blockages and could be a valuable screening tool for detecting coronary artery obstructions, potentially enabling their use in routine health checks and in identifying patients at high risk for future coronary events.

Author 1: Mohd Syazwan Md Yid

Author 2: Rosmina Jaafar

Author 3: Noor Hasmiza Harun

Author 4: Mohd Zubir Suboh

Author 5: Mohd Shawal Faizal Mohamad

Keywords: Deep learning; CNN; LSTM; ATTN; simultaneous ECG and PPG; coronary artery disease

Paper 75: Increasing the Performance of Iceberg Query Through Summary Tables

Abstract: One of the key challenging problems in data mining is data retrieval from large data repositories, as the sizes of data are growing very fast, to deal with this situation, there is a need for efficient data mining techniques. For efficient mining tasks number of queries have been emerged. Iceberg query is one of them, in which the output is much smaller like the tip of the iceberg as compared to the large input dataset, these queries take very long processing time and require a huge amount of main memory. However the processing devices have limited memories, so the efficient processing of iceberg queries is a challenging problem for most of the researchers. In this paper we present a novel technique, namely a summary table, to address this problem. Specifically, we adopt the summary table technique to acquire the required results at summary levels. The experimental results demonstrate that the summary table technique is highly effective for large datasets. Compared to bitmap indexing and cubed techniques, the summary table offers faster retrieval capabilities. Furthermore, the proposed technique achieved state-of-the-art performance.

Author 1: Gohar Rahman

Author 2: Wajid Ali

Author 3: Mehmood Ahmed

Author 4: Hassan Jamil Sayed

Author 5: Mohammad A. Saleh

Keywords: Threshold (TH); bitmap index; aggregate function; Iceberg Query (IB); anti-monotone; non-anti-monotone aggregation

Paper 76: Enhancing Supply Chain Transparency and Efficiency Through Innovative Blockchain Solutions for Optimal Operations Management

Abstract: Blockchain technology holds the potential to revolutionize supply chain management by ensuring transparency, efficiency, and security. This paper presents a detailed examination of blockchain's implementation in supply chain systems, focusing on safeguarding confidential information and preserving supply chain integrity. The method involves extracting ‘sales order’ data from Walmart’s transactional database, which is then encrypted using AES algorithms to protect sensitive details such as client names and geographical information. Utilizing Ethereum's decentralized architecture, smart contracts are employed to manage transactions, encryption, decryption, and access rights. The Ethereum P2P network also aids in data validation and asset preservation, enhancing the system’s reliability. Comparative analysis shows that the proposed encryption method, with encryption and decryption times of 2.8 and 3.2 seconds, outperforms traditional methods like RSA and ABE. Implemented in Python, this blockchain-based technique offers a robust, nearly infallible solution that can be applied to various supply chain practices, including Asset Management (AM), Enterprise Asset Management (EAM), and Supply Chain Management (SCM), addressing contemporary challenges and enhancing operational efficiency.

Author 1: Shamrao Parashram Ghodake

Author 2: Vishal M. Tidake

Author 3: Sanjit Singh

Author 4: Elangovan Muniyandy

Author 5: Mohit

Author 6: Lakshmana Phaneendra Maguluri

Author 7: John T Mesia Dhas

Keywords: Blockchain; supply chain management; advanced encryption standard; Ethereum blockchain; data storage

Paper 77: Enhanced Quantitative Financial Analysis Using CNN-LSTM Cross-Stitch Hybrid Networks for Feature Integration

Abstract: This research paper provides innovative approaches to support financial prediction, or it is a different kind of economic prediction that extends over collecting different economic information. Financial prediction is a concept that has been employed. The present study offers a unique approach to predicting finances by integrating many financial issues utilizing a cross-stitch hybrid approach. The method uses information from several financial databases, including market data, corporate reports, and macroeconomic indicators, to create a comprehensive dataset. Employing MinMax normalization the features are equally scaled to provide uniform input for the algorithm. The combination of Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) systems form the basis of the framework. To understand the time-dependent nature of financial information, LSTM networks (long short-term memory) are utilized to record and simulate the temporal interactions and patterns. Concurrently, spatial features are extracted using CNNs; these components help identify patterns that are difficult to identify with conventional techniques. Better handling of risks, more optimal approaches to investing, and more informed decision-making are made possible by the enhanced forecasting potential that this method—which is described above—offers. Potential pilot studies will focus on innovative uses in financial decision-making and advancements in cross-stitching structure. This paper proposes a sophisticated approach that can help stakeholders, such as investors, analysts of data, and other financial intermediaries, traverse the complexities of financial markets.

Author 1: Taviti Naidu Gongada

Author 2: B. Kumar Babu

Author 3: Janjhyam Venkata Naga Ramesh

Author 4: P. N. V. Syamala Rao M

Author 5: K. Aanandha Saravanan

Author 6: K Swetha

Author 7: Mano Ashish Tripathi

Keywords: Cross-Stitch Hybrid Networks; predictive modelling; LSTM networks; convolutional neural networks; financial analysis

Paper 78: Enhanced Early Detection of Oral Squamous Cell Carcinoma via Transfer Learning and Ensemble Deep Learning on Histopathological Images

Abstract: Oral Squamous Cell Carcinoma (OSCC) is one main kind of oral cancer; early diagnosis is rather important to increase patient survival chances. This study investigates the application of advanced deep learning techniques including transfer learning and ensemble learning to increase the accuracy of oral squamous cell cancer (OSCC) diagnosis using histopathological image analysis. Two transfer learning models, EfficientNetB3 and ResNet50, support the suggested method to extract suitable features from the histopathological images. Both models permit fine-tuning to improve their classification accuracy. On tests taken after the initial training, the EfficientNetB3 model scored 96.15%. Later on, training ResNet50 yielded a test accuracy of 91.40%. Weighted voting merged several models into an ensemble model designed to maximize the strengths of each network. With a test accuracy of 98.59% and a training accuracy of 99.34%, the ensemble model showed notably higher performance than the values obtained by the individual models. Divided into OSCC and standard categories, the collection has 5,192 extremely well-resolved images. The images were used to create training, validation, and testing sets. We used this method to consistently evaluate the model's performance and reduce overfitting. Furthermore, the ensemble model proved to be quite accurate with recall and F1 scoring, thereby proving its capacity to routinely identify OSCC images. Both groups produced ROC curves, and the area under the curve (AUC) demonstrated excellent model performance. Transfer learning and ensemble learning are used together in this study to show that OSCC can be found early and consistently in histopathology images. The findings reveal that the recommended strategy could be a consistent tool to assist pathologists in the precise and timely detection of OSCC, thereby improving patient treatment and outcomes.

Author 1: Gurjot Kaur

Author 2: Sheifali Gupta

Author 3: Ashraf Osman Ibrahim

Author 4: Salil bharany

Author 5: Marwa Anwar Ibrahim Elghazawy

Author 6: Hadia Abdelgader Osman

Author 7: Ali Ahmed

Keywords: Oral Squamous Cell Carcinoma (OSCC); histopathology images; transfer learning; ensemble learning; EfficientNetB3; ResNet50; deep learning; cancer detection; medical image analysis

Paper 79: Furniture Panel Processing Positioning Design Based on 3D Measurement and Depth Image Technology

Abstract: In recent years, furniture panel processing positioning based on computer vision technology has received increasing attention. A 3D measurement imaging technology based on laser scanning technology is proposed to address the significant environmental impact of traditional visual technology. Subsequently, deep image processing techniques are introduced to address the high image noise. In the experiment of measuring panel using 3D measurement technology, 14 measurement lines were taken every 10mm of the measurement length. The maximum measurement value was 204.62mm, the minimum measurement value was 204.37mm, and the manual measurement result was 204.5mm. 14 measurement lines were taken every 14mm of the measurement length. The maximum measurement value was 134.15mm, the minimum measurement value was 133.894mm, and the manual measurement was 134.1mm. 14 measurement lines were taken every 14mm of the measurement thickness. The maximum measurement value was 26.646mm, the minimum measurement value was 26.242mm, and the result of manual measurement was 26.5mm. The 3D imaging technology based on laser scanning is relatively accurate in measuring the 3D data of panels, which can be applied in the positioning and detection system of panel processing. In addition, the experiment compares depth image processing methods, verifying the effectiveness of the designed method. Meanwhile, this research also has certain reference significance for exploring the real-time positioning of other objects.

Author 1: Binglu Chen

Author 2: Guanyu Chen

Author 3: Qianqian Hu

Keywords: Laser scanning; 3D measurement; deep image processing; positioning; computer vision

Paper 80: Research and Implementation of Facial Expression Recognition Algorithm Based on Machine Learning

Abstract: Traditional information security management methods can provide a degree of personal information protection but remain vulnerable to issues such as data breaches and password theft. To bolster information security, facial expression recognition offers a promising alternative. To achieve efficient and accurate facial expression recognition, we propose a lightweight neural network algorithm called T-SNet (Teacher-Student Net). In our approach, the teacher model is an enhanced version of ResNet18, incorporating fine-grained feature extraction modules and pre-trained on the MS-Celeb-1M facial dataset. The student model uses the lightweight convolutional neural network ShuffleNetV2, with the model's accuracy further improved by optimizing the distillation loss function. This design carefully considers the key features of facial expressions, determines the most effective extraction techniques, and classifies and recognizes these features. To evaluate the performance of our algorithm, we conducted comparative experiments against state-of-the-art facial expression recognition methods. The results show that our approach outperforms existing methods in both recognition accuracy and efficiency.

Author 1: Xinjiu Xie

Author 2: Jinxue Huang

Keywords: Facial expression; expression recognition; convolutional neural network; deep learning

Paper 81: Enhancing Music Emotion Classification Using Multi-Feature Approach

Abstract: Emotions are a fundamental aspect of human expression, and music lyrics are a rich source of emotional content. Understanding the emotions conveyed in lyrics is crucial for a variety of applications, including music recommendation systems, emotion classification, and emotion-driven music composition. While extensive research has been conducted on emotion classification using audio or combined audio-lyrics data, relatively few studies focus exclusively on lyrics. This gap highlights the need for more focused research on lyric-based emotion classification to better understand its unique challenges and potentials. This paper introduces a novel approach for emotion classification in music lyrics, leveraging a combination of natural language processing (NLP) techniques and dimension reduction methods. Our methodology systematically extracts and represents the emotional features embedded within the lyrics, utilizing a diverse set of NLP techniques and integrating new features derived from various emotion lexicons and text analysis. Through extensive experimentation, we demonstrate the effectiveness of our approach, achieving significant improvements in accurately classifying the emotions expressed in music lyrics. This study underscores the potential of lyric-based emotion analysis and provides a robust framework for further research in this area.

Author 1: Affreen Ara

Author 2: Rekha V

Keywords: Emotion classification; music lyrics; feature extraction; lexicon features

Paper 82: Art Image Color Sequence Data Processing Method Based on Artificial Intelligence Technology

Abstract: With the traditional quality enhancement methods cannot control the best field density range resulting in too large threshold value of colour difference in art works. Therefore, a research on art works quality enhancement based on image processing technology is proposed. The CIE L* a* b* color space model is established to divide the color magnitude and then transform the color space by RGB space conversion model. On this basis, the quality of art works is enhanced according to the process of the quality enhancement of art works. As considering that the actual density is not within the control range, the image processing technology is used to separate targets to solve this problem. In the experiment, Adobe Illustrator CS6 software was used to make the experimental color target and six test samples were selected to test whether the distribution results of the two methods in different degree of color difference perception met the quality enhancement requirements. The experimental results show that the quality enhancement effect of the proposed method is better and more in line with the design requirements.

Author 1: Xujing Zhao

Author 2: Xiwen Chen

Author 3: Jianfei Shen

Keywords: Artworks; quality enhancement; image processing technology; color space model; target separation; experimental color target

Paper 83: AI-Driven Prioritization Techniques of Requirements in Agile Methodologies: A Systematic Literature Review

Abstract: Software requirements are the foundation of a successful software development project, outlining the customer's expectations for the software's functionality. Conventional techniques of requirement prioritization present several challenges, such as scalability, customer satisfaction, efficiency, and dependency management. These challenges make the process difficult to manage effectively. Prioritizing requirements by setting criteria in order of importance is essential to addressing these issues and ensuring the efficient use of resources, especially as software becomes more complex. Artificial intelligence (AI) offers promising solutions to these challenges through algorithms like Machine Learning, Fuzzy Logic, Optimization, and Natural Language Processing. Despite the availability of reviews on conventional prioritization techniques, there is a notable gap in comprehensive reviews of AI-based methods. This paper offers a systematic literature review (SLR) of AI-driven requirements prioritization techniques within Agile methodologies, covering 32 papers published between 2010 and 2024. We conducted a parametric analysis of these techniques, identifying key parameters related to both the prioritization process and specific AI methods. Our findings clarify the application domains of various AI-based techniques, offering crucial insights for researchers, requirement analysts, and stakeholders to choose the most effective prioritization methods. These insights consider dependencies and emphasize the importance of collaboration between stakeholders and the development team for optimal results.

Author 1: Aya M. Radwan

Author 2: Manal A. Abdel-Fattah

Author 3: Wael Mohamed

Keywords: Requirement analysis; requirement prioritization; agile; fuzzy logic; machine learning; optimization

Paper 84: Intelligent Detection and Search Model for Communication Signals Based on Deep-Re-Hash Retrieval Technology

Abstract: With the explosive growth of image data, traditional image retrieval methods face challenges of low efficiency and insufficient accuracy. In view of this, the study first analyzed the traditional Deep-Re-Hash detection technology and constructed a general hash detection model. Secondly, Cauchy functions and Hadamard matrices were introduced to optimize the generation of hash centers, and an improved Deep-Re-Hash detection model was proposed. The experimental results showed that the highest precision of the improved Deep-Re-Hash was 97%, and the highest MAP value was 90%. In simulation testing, the lowest detection similarity of the improved Deep-Re-Hash detection model was 64.8%, and the detection speed at this time was 7.6s. The hash codes generated by this model were highly aggregated, with very clear edges. In the indicator rating, the highest storage occupancy rating was close to 45 points, the highest detection satisfaction rating was close to 50 points, and the highest detection time rating was close to 30 points. Based on the above data, the proposed improved Deep-Re-Hash detection model shows great potential in processing large-scale image data. It successfully improves the intelligent detection and search efficiency of communication image signals, providing useful reference and inspiration for researchers in related fields.

Author 1: Hui Liu

Author 2: Xupeng Liu

Keywords: Deep-Re-Hash retrieval; communication signals; image data; cauchy function; hadamard matrix

Paper 85: Content-Based Image Retrieval Using Transfer Learning and Vector Database

Abstract: Content-based image retrieval (CBIR) systems are essential for efficiently searching large image datasets using image features instead of text annotations. Major challenges include extracting effective feature representations to improve accuracy, as well as indexing them to improve the retrieval speed. The use of pre-trained deep learning models to extract features has elicited interest from researchers. In addition, the emergence of open-source vector databases allows efficient vector indexing which significantly increases the speed of similarity search. This paper introduces a novel CBIR system that combines transfer learning with vector databases to improve retrieval speed and accuracy. Using a pre-trained VGG-16 model, we extract high-dimensional feature vectors from images, which are stored and retrieved using the Milvus vector database. Our approach significantly reduces retrieval time, achieving real-time responses while maintaining high precision and recall. Experiments conducted on ImageClef, ImageNet, and Corel-1k datasets demonstrate the system’s effectiveness in large-scale image retrieval tasks, outperforming traditional methods in both speed and accuracy.

Author 1: Li Shuo

Author 2: Lilly Suriani Affendey

Author 3: Fatimah Sidi

Keywords: Content-based image retrieval (CBIR); image retrieval; transfer learning; convolutional neural networks; VGG-16; vector database; milvus; feature extraction; high-dimensional vectors; real-time image search

Paper 86: Towards Accurate Detection of Diabetic Retinopathy Using Image Processing and Deep Learning

Abstract: Diabetic retinopathy (DR) is a critical complication of diabetes, characterized by pathological changes in retinal blood vessels. This paper presents an innovative software application designed for DR detection and staging using fundus images. The system generates comprehensive reports, facilitating treatment planning and improving patient outcomes. Our study aims to develop an affordable computer assisted analysis system for accurate DR assessment, leveraging publicly available fundus image datasets. Key objectives include identifying relevant features for DR staging, developing robust image processing algorithms for lesion detection, and implementing machine learning models for accurate diagnosis. The research employs various pre-processing techniques to enhance image quality and optimize feature extraction. Convolutional Neural Networks (CNNs) are utilized for stage classification, achieving an impressive accuracy of 93.45%. Lesion detection algorithms, including optic disk localization, blood vessel segmentation, and exudate identification, demonstrate promising results in accurately identifying DR-related abnormalities. The developed software product integrates these advancements, providing a user-friendly interface for efficient DR diagnosis and management. Evaluation results validate the effectiveness of the CNN model in stage classification and lesion detection, with high sensitivity and specificity. The study discusses the significance of image augmentation and hyperparameter tuning in improving model performance. Future research directions include enhancing the detection of microaneurysms and hemorrhages, incorporating higher resolution images, and standardizing evaluation methods for lesion detection algorithms. In conclusion, this research underscores the potential of technology in revolutionizing DR diagnosis and management. The developed software product offers a cost-effective solution for early DR detection, emphasizing the importance of accessible healthcare solutions. The findings contribute to advancing the field of DR analysis and inspire further innovation for improved patient care.

Author 1: K. Kalindhu Navanjana De Silva

Author 2: T. Sanduni Kumari Lanka Fernando

Author 3: L. D. Lakshan Sandaruwan Jayasinghe

Author 4: M.H.Dinuka Sandaruwan Jayalath

Author 5: Kasun Karunanayake

Author 6: B.A.P. Madhuwantha

Keywords: Diabetic retinopathy; fundus images; computer-assisted analysis; deep learning; image processing; convolutional neural networks component

Paper 87: Machine Learning Approaches for Predicting Occupancy Patterns and its Influence on Indoor Air Quality in Office Environments

Abstract: It is normal for the modern population to spend 12 hours or more daily indoors where the level of comfort can be moderated. Yet, indoor occupants are similarly exposed to various air pollutants just as outdoors. Indoor air pollution could be detrimental toward the occupant's health noted by the United Nation Environment Programme (UNEP) in the Pollution Action Note, published on 7th of September 2021. According to the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) standards, occupancy patterns could influence indoor air quality. Hence, this paper investigates the utilisation of machine learning algorithms in predicting occupancy patterns against indoor air quality (IAQ) variables such as humidity, temperature, light, and carbon dioxide (CO2). This study compares the performance of selected machine learning approaches, namely deep learning (LSTM, CNN), regression (ANN) and (SVR) models. In addition, it explores the diverse range of evaluation metrics utilized to evaluate the performance of machine learning in the specific context of Mean Squared Error (MSE) and Mean Absolute Error (MAE). In the training phase, the SVR model achieved the lowest MAE of 0.0826 and MSE of 0.0280 as compared to the other algorithms. The ANN model demonstrated slightly better generalization capabilities in the testing phase, while the LSTM model demonstrated robust performance in the test phase. Overall, the results highlighted the significant impact of occupancy behaviour on Indoor Air Quality (IAQ) variables and underscored the importance of advanced modelling techniques in IAQ monitoring and management, emphasizing the need for tailored approaches to address the complex relationship between occupancy patterns and IAQ variables.

Author 1: Amir Hamzah Mohd Shaberi

Author 2: Sumayyah Dzulkifly

Author 3: Wang Shir Li

Author 4: Yona Falinie A. Gaus

Keywords: Indoor air quality; occupancy patterns; machine learning; deep learning; regression models; Mean Squared Error; Mean Absolute Error; IAQ monitoring; IAQ management

Paper 88: Visualization of Personality and Phobia Type Clustering with GMM and Spectral

Abstract: Personality traits, the unique characteristics defining individuals, have intrigued philosophers and scholars for centuries. With recent advances in machine learning, there is an opportunity to revolutionize how we understand and differentiate personality traits. This study seeks to develop a robust cluster analysis approach (unsupervised learning) to efficiently and accurately classify individuals based on their personality traits, overcoming the limitations of manual classification. The problem at hand is to create a system that can handle the subjective nature of qualitative personality data, providing insights into how people interact, collaborate, and behave in various social contexts and thus increase learning opportunities. To achieve this, various unsupervised clustering techniques, including spectral clustering and Gaussian mixture models, will be employed to identify similarities in unlabeled data collected through interview questions. The clustering approach is crucial in helping policy makers to identify suitable approaches to improve teamwork efficiency in both educational institutions and job industries.

Author 1: Ting Tin Tin

Author 2: Cheok Jia Wei

Author 3: Ong Tzi Min

Author 4: Lim Siew Mooi

Author 5: Lee Kuok Tiung

Author 6: Ali Aitizaz

Author 7: Chaw Jun Kit

Author 8: Ayodeji Olalekan Salau

Keywords: Unsupervised learning; learning opportunities; clustering; personality; machine learning; Gaussian mixture model; spectral clustering

Paper 89: A Comprehensive Study of BIM for Infrastructural Crack Detection and the Vital Strategies

Abstract: Building information modelling is one of the emerging technologies in the construction industry and is relevant to its productivity and efficiency. Application which affects the product and process of the industry. An underdeveloped area with less attention is its adoption for crack detection and visualisation for infrastructural maintenance. This study provides a thorough perspective on BIM adoption for crack detection and visualisation. It also identified the different strategies that can aid in the adoption and use of BIM for infrastructural monitoring and maintenance in South Africa. The study adopted a quantitative approach, and questionnaires were distributed to industry professionals through an online platform. The collected data was analysed. The results indicate a need for incorporation of this aspect into the HEI curriculum and a teaching approach that is practical and experimental to be adopted.

Author 1: Samuel Adekunle

Author 2: Opeoluwa Akinradewo

Author 3: Babatunde Ogunbayo

Author 4: Andrew Ebekozien

Author 5: Clinton Aigbavboa

Keywords: Developing country; emerging technology; facility management; visualisation; emerging technology; South Africa

Paper 90: The Adoption of Electronic Payments in Online Shopping: The Mediating Role of Customer Trust

Abstract: This study investigates the factors influencing electronic payment in online shopping behavior among Ho Chi Minh City consumers. With the rapid advancement of technology, e-commerce has become a new trend, and understanding the intention to adopt electronic payment is crucial for online businesses. The research employs quantitative and qualitative methods, utilizing a survey of 437 Ho Chi Minh City consumers. The data collected is processed using SPSS 24 and SmartPls4 software. Eight factors related to consumers' intention to use electronic payment are identified: social influence, security, perceived usefulness, convenience, ease of use, customer trust, perceived risk, and performance expectancy. The study's findings will contribute to the existing knowledge base for businesses, facilitating the promotion of electronic payment adoption. This support will aid businesses in developing more attractive online sales strategies, encouraging consumers to shop and pay online more frequently and, at the same time, contribute to supporting departments in formulating policies for digital payments, thereby promoting national digital transformation.

Author 1: Nguyen Thi Phuong Giang

Author 2: Thai Dong Tan

Author 3: Le Huu Hung

Author 4: Nguyen Binh Phuong Duy

Keywords: Electronic payment; Intention to use; Online shopping

Paper 91: Moving Beyond Traditional Incident Response: Combating APTs with Warfare-Enabled Continuous Response

Abstract: Critically examining the cybersecurity management practices, it can be concluded that security management used by the organizations is mostly control-centered against a wide range of threats to information systems. This control-centered approach has matured to act as a shield to prevent against a large variety of attacks. Since threats against the information systems are becoming sophisticated, persistent and evolving, therefore, the current approach has not been very effective against the advanced strategies and techniques used by the emerging threats like APTs (Advanced Persistent Threats). The core argument of this paper suggests that to match up the capabilities of APTs, organizations need a major shift in their strategies. This shift needs to focus more on the response oriented techniques relegating erstwhile prevention-centered approach. Traditionally the warfare strategies are more response oriented. Some of the non-kinetic strategies (not involving physical fighting) can be useful in developing response capability of Information Systems. Therefore, drawing on the warfare paradigm, and making use of DCT (Dynamic Capability Theory), this research examines the applicability of warfare strategies in the entrepreneur domain. This article will also contribute by means of a research framework arguing that the integration of prevailing information security capabilities; such as incident response capabilities and security capabilities from the warfare practices is possible resulting in dynamic capabilities (warfare-enabled). Such capabilities can improve security performance.

Author 1: Abid Hussain Shah

Keywords: Information operations; information warfare; cyber security; dynamic capabilities; incident response capabilities; warfare enabled capabilities

Paper 92: Deep Learning-Based Image Recognition Technology for Wind Turbine Blade Surface Defects

Abstract: This paper proposes WindDefectNet, an image recognition system for surface defects of wind turbine blades, aiming at solving the key problems in wind turbine blade maintenance. At the beginning of the system design, the functional requirements and performance index requirements are clarified to ensure the realization of the functions of image acquisition and preprocessing, defect detection and classification, defect localization and size measurement, and to emphasize the key performance indexes such as accuracy, recall, processing speed and robustness of the system. The system architecture consists of multiple modules, including image acquisition and preprocessing module, feature extraction module, attention enhancement module, defect detection module, etc., which work together to achieve efficient defect recognition and localization. By adopting advanced deep learning techniques and model design, WindDefectNet is able to maintain high accuracy and stability in complex environments. Experimental results show that WindDefectNet performs well under different lighting conditions, shooting angles, wind speed and weather conditions, and has good environmental adaptability and robustness. The system provides strong technical support for blade maintenance in the wind power industry.

Author 1: Zheng Cao

Author 2: Qianming Wang

Keywords: Wind turbine blades; image recognition; defect detection; deep learning; WindDefectNet

Paper 93: Design of Intelligent Extraction Method for Key Electronic Information Based on Neural Networks

Abstract: With the rapid development of the Internet and other emerging media, how to find the needed information from massive electronic documents in time and accurately has become an urgent problem. A key electronic information extraction method based on neural network learning ideas has been proposed to solve the problems of time-consuming and difficult deep semantic feature mining in traditional text classification methods. Firstly, a weighted graph model was introduced to improve the TextRank keyword extraction algorithm, helping to capture complex data information and implicit semantics. The results indicate that the optimization method has the highest extraction accuracy (96.52%) on the CSL dataset, and its performance in feature extraction of information data is superior to other comparative models. Secondly, combining LSTM and self attention mechanism to achieve key feature extraction of contextual semantic information. The results indicate that this optimization method has relatively small training and testing errors in data classification, and tends to converge in the later stages of iteration. The accuracy of information extraction reached 94.37%, which is better than other comparative models. The keyword extraction integrity of the fusion model on the THUCNews dataset and Sogou News dataset were 86.2 and 84.1, respectively, with consistency of 96.3 and 94.7, and grammatical correctness of 92.1 and 92.2, respectively. The neural network-based extraction method proposed by the research institute can not only effectively improve the accuracy of information extraction, but also adapt to the changing data environment, and has great potential for application in the field of electronic information processing.

Author 1: Xiaoqin Chen

Author 2: Xiaojun Cheng

Keywords: Key electronic information; intelligent extraction; TextRank; LSTM; context

Paper 94: Optimized Blockchain-Based Deep Learning Model for Cloud Intrusion Detection

Abstract: Cyberattacks are becoming increasingly complex and subtle. In many different types of networks, intrusion detection systems, or IDSs, are frequently employed to help in the prompt detection of intrusions. Blockchain technology has gained a lot of attention recently as a means of sharing data without a reliable third party. Specifically, it is impossible to change data stored in one block without changing all the following blocks. Create a deep learning (DL) method based on blockchain technology and hybrid optimization to improve the IDS's prediction accuracy. The UNSW-NB15 dataset is gathered via the Kaggle platform and utilized for Python system training. Principal component analysis (PCA) is used in the preprocessing to eliminate errors and duplication. Next, employ association rule learning (ARL) and information gain (IG) approaches to retrieve pertinent characteristics. The greatest features are the ones that improve detection performance through hybrid seahorse and bat optimization (HSHBA) selection. Lastly, create an efficient intrusion detection system by designing Blockchain-based Ensemble DL (BEDL) models, with convolutional neural networks (CNNs), restricted Boltzmann machines (RBM), and generative adversarial networks (GAN). The constructed model's experimental results are verified using pre-existing classifiers, yielding an improved accuracy of 99.12% and precision of 99%.

Author 1: Sultan Alasmari

Keywords: Intrusion detection system; blockchain; deep learning; hybrid optimization; cloud computing; feature selection

Paper 95: Deep Learning for Stock Price Prediction and Portfolio Optimization

Abstract: Using deep learning for stock market predictions and portfolio optimizations is a burgeoning field of research. This study focuses on the stock market dynamics in developing countries, which are often considered less stable than their developed counterparts. The study is structured in two stages. In the first stage, the authors introduce a stacked LSTM model for predicting NIFTY stocks and then rank the stocks based on their predicted returns. In the second stage, the high-return stocks are selected to form 30 different portfolios with six different objectives, each comprising the top 7, 8, 9, and 10 NIFTY stocks. These portfolios are then compared based on risk and returns. Experimental results show that portfolios with five stocks offer the best returns and that adding more than nine stocks to the portfolio leads to excessive diversification and complexity. Therefore, the findings suggest that the proposed two-stage portfolio optimization method has the potential to construct a promising investment strategy, offering a balance between historical and future information on assets.

Author 1: Ashy Sebastian

Author 2: Veerta Tantia

Keywords: Deep learning; long-short term memory; stock price prediction; portfolio optimization; emerging markets; Indian stock market

Paper 96: An Efficient Hierarchical Mechanism for Handling Network Partitioning Over Mobile Ad Hoc Networks

Abstract: Mobile ad hoc networks exhibit distinctive challenges e.g., limited transmission range and dynamic mobility of the participating nodes. These challenges serve as the reasons for the frequent occurrence of network partitioning in mobile ad hoc networks. Network partitioning happens when a linked network topology is partitioned into two or more independent partitions. Because of this phenomenon, the participating node in one partition maintains no linkage with a node in another partition. Network partitioning results in the inaccessibility of mapping knowledge, logical labeling space, and logical structure of the participating nodes. As a result, the performance of a distributed hash tables (DHTs)-oriented routing mechanism is severely affected. In DHT-oriented routing methodologies, the logical network identifier of a new participating node is calculated by considering the logical network identifiers of all the physical neighboring nodes. The logical network identifiers are utilized for routing of packets from a source participating node to a destination participating node in the network. In the event of network partitioning, the incorrect computation of logical network identifiers happens concerning the physical proximity of the participating nodes. This research work suggests an effective routing mechanism to deal with the aforementioned network partitioning-related issues. Simulation results prove the superiority of the suggested scheme over the existing mechanisms.

Author 1: Ali Tahir

Author 2: Fathe Jeribi

Keywords: Mobile Ad Hoc networks; network partitioning; distributed hash tables; logical cluster member node; logical cluster leader; logical network identifier

Paper 97: A Proposed λ_Mining Model for Hierarchical Multi-Level Predictive Recommendations

Abstract: Delivering the most suitable products and services essentially relies on successfully exploring the potential relationship between customers and products. This immense need for intelligent exploration has led to the emergence of recommendation systems. In an environment where an immense variety exists, it is vital for buyers to own an intelligent exploratory map to guide them in finding their choices. Personalization has proven to be a successful contributor to recommenders. It provides an accurate guide to explore the users’ preferences. In the field of recommendation systems, the performance of the systems has been continuously measured by their success in accurate, personalized recommendations. There is no argument that personalization is one key success; however, this research argues that recommendation systems are not only about personalization. Other success factors should be considered in targeting optimality. The current research explores the hierarchy map representing the strengths and dependencies of the recommendation systems pillars associated with their influence level and relationships. Moreover, the research proposes a novel predictive approach that applies a hybrid of content and collaborative filtering recommendation systems to provide the most suitable customer recommendations effectively. The model utilizes a proposed features selection approach to detect the most significant features and explore the most effective associations’ schemes for the recommendations label feature. The proposed model is validated using a benchmark dataset by extracting direct and transitive associations and following the identified schematic for the required recommendations. The classification techniques are applied, proving the model's applicability with an accuracy ranging from 96% to 99%.

Author 1: Yousef S. Alsahafi

Author 2: Ayman E. Khedr

Author 3: Amira M. Idrees

Keywords: Recommendation systems; data mining; features selection; associations rules mining; classification techniques

Paper 98: Proposal of OptDG Algorithm for Solving the Knapsack Problem

Abstract: In a computational complexity theory, P, NP, NP-complete and NP-hard problems are divided into complexity classes which are used to emphasize how challenging it is to solve particular types of problems. The Knapsack problem is a well-known computational complexity theory and fundamental NP-hard optimization problem that has applications in a variety of disciplines. Being one of the most well-known NP-hard problems, it has been studied extensively in science and practice from theoretical and practical perspectives. One of the solution to the Knapsack problem is the Dantzig’s greedy algorithm which can be expressed as a linear programming algorithm which seeks to discover the optimal solution to the knapsack problem. In this paper, an optimized Dantzig greedy (OptDG) algorithm that addresses frequent edge cases, is suggested. Furthermore, OptDG algorithm is compared with the Dantzig’s greedy and optimal dynamically programmed algorithms for solving the Knapsack problem and performance evaluation is conducted.

Author 1: Matej Arlovic

Author 2: Tomislav Rudec

Author 3: Josip Miletic

Author 4: Josip Balen

Keywords: Dynamic programming; Dantzig algorithm; greedy algorithm; knapsack problem; linear programming; NP-Problem; optimization; OptDG

Paper 99: A Relevant Feature Identification Approach to Detect APTs in HTTPS Traffic

Abstract: This study addresses the significant challenges posed by Advanced Persistent Threats (APTs) in modern computer networks, particularly their use of DNS to establish covert communication via command and control (C&C) servers. The advent of TLS 1.3 encryption further complicates detection efforts, as critical data within DNS over HTTPS (DoH) traffic remains inaccessible, and decryption would compromise user privacy. APTs frequently leverage Domain Generation Algorithms (DGAs), necessitating real-time detection solutions based on immediately accessible features within HTTPS traffic. Current research predominantly focuses on system-level behavioral analysis, often neglecting the proactive potential offered by Cyber Threat Intelligence (CTI), which can reveal malicious patterns through Techniques, Tactics, and Procedures (TTPs) and Indicators of Compromise (IoCs). This study proposes an innovative approach utilizing the MITRE ATT&CK framework to identify relevant features in the face of encryption and the inherent complexity of APT activities. The primary objective is to develop a robust dataset and methodology capable of detecting APT behaviors throughout their lifecycle, emphasizing a lightweight, cost-effective solution through passive monitoring of network traffic to ensure real-time detection. The key contributions of this research include an in-depth analysis of the encryption challenges in detecting DNS-based APTs, a thorough examination of APT attack strategies using DNS, and the integration of CTI to enhance detection capabilities. Moreover, this study introduces the KAPT 2024 dataset, generated by the KExtractor tool, and demonstrates the effectiveness of the detection model through experiments with a variety of machine learning algorithms. The results underscore the potential for this approach to significantly improve APT detection in encrypted network environments.

Author 1: Abdou Romaric Tapsoba

Author 2: Tounwendyam Frederic Ouedraogo

Keywords: DNS over HTTPS; advanced persistent threats; machine learning; cyber threat intelligence; MITRE ATT&CK; domain generation algorithms

Paper 100: Multiclass Fruit Detection Using Improved YOLOv3 Algorithm

Abstract: Manual interventions continue to be used in fruit-picking and billing at large-scale fruit storage facilities. Recent advances in deep in learning approaches, such as one-stage detectors like You Only Look Once (YOLO) and Single Stage Detector (SSD), as well as two-stage detectors like Faster RCNN and Mask RCNN, aim to streamline the processes involved with fruit detection and enhance efficiency. However, these frameworks continue to suffer with multi-scale objects, in terms of performance and efficiency due to large parameter sizes. These problems increase when multi-class fruits are encountered. We propose an improved version of the one-stage detector framework YOLOv3 for multi-class fruit detection. Our proposed model addresses the challenges of multi-scale object detection and detection of different fruit types in an image by incorporating CNN, bottleneck, and Spatial Pyramid Pooling Fast (SPPF) modules in the Head, Neck, and custom backbone of the YOLOv3 framework. Optimization of learnable parameters for computational efficiency is achieved by concatenating features at different feature map resolutions. The proposed model incorporates fewer layers and parameters compared to YOLOv3 and YOLOv5 models. We performed extensive testing on three datasets downloaded from Roboflow and compared them with YOLOv3 and YOLOv5 models. Our model achieved mAP50 of 0.747 on Dataset 1 comprising images of apples, bananas, and oranges whereas Dataset 2 consisting of images of apples, oranges, lemon, and Pear, achieved mAP50 of 0.981. Testing the Mineapple dataset comprising on-tree images of apples of varied sizes, achieved an accuracy of 0.643. We observe that the performance of our model beats the performance of the YOLOv3 and YOLOv5 models.

Author 1: Seema C. Shrawne

Author 2: Jay Sawant

Author 3: Omkar Chaubal

Author 4: Karan Suryawanshi

Author 5: Diven Sirwani

Author 6: Vijay K. Sambhe

Keywords: Precision agriculture; yield estimation; fruit detection; YOLOv3; feature concatenation; spatial contexting

Paper 101: A Meta-Heuristics-Based Solution for Multi-Objective Workflow Scheduling in Fog Computing

Abstract: In recent years, there has been a significant increase in the volume of data generated by Internet of Things (IoT) applications, mostly driven by the rapid proliferation of IoT devices and advancements in communication technologies. The conventional cloud computing network was not specifically built to handle such a vast volume of data, leading to several issues, including increased processing time, higher costs, larger band-width usage, increased power usage, and communication delays. As a solution, conventional cloud servers have been expanded to additional layers of computing, storage, and network, termed as cloud-fog computing. The cloud-fog computing provides storage, processing, networking, and analytics capabilities in close proximity to IoT devices. The problem of scheduling work-flow applications in cloud-fog environments to optimize several conflicting objectives is classified as computationally complex. Particle Swarm Optimization is the widely recognized evolutionary meta-heuristic and is the optimal method for implementing multi-objective solutions because of its user-friendly approach and quick converging capability. Despite its wide acceptance, it does have several drawbacks, such as early convergence and solution stagnation. In order to overcome these limitations, this paper establishes a comprehensive theoretical model to schedule workflow applications for cloud-fog systems. The proposed model employs various competing objectives, such as power usage, overall cost, and makespan. To achieve this, we introduce a novel algorithm, learning enhanced particle swarm optimization (LE-PSO), which incorporates an inverse tangent inertia weight policy and adaptive learning factor methods. The efficiency of the LE-PSO is subsequently assessed by employing an operational data set of scientific workflow applications within a cloudsim-based simulation and validated against GAMPSO, EMMOO, PSO, and GA state-of-the-art approaches. The workflow scheduling, we suggest achieves the substantial decrease in makespan and power usage while maintaining the total cost at an optimal level, in comparison to existing meta-heuristics.

Author 1: Gyan Singh

Author 2: Vivek Dubey

Keywords: Fog computing; DAG; workflow applications; makespan; energy; PSO

Paper 102: Intelligent Control Technology of Electric Pressurization Based on Fuzzy Neural Network PID

Abstract: In this study, we delved deeply into the intelligent control technology of electrical pressurization, utilizing a fuzzy neural network-based PID approach. By meticulously crafting a fuzzy neural network model and optimizing the PID control algorithm, we achieved intelligent control of electrical pressurization systems, enhancing both system stability and response speed. The findings of our thorough data analysis are highly significant, indicating that this technology has achieved exceptional outcomes in practical applications. This paper delves into a comparative analysis of the performance between intelligent electrical pressurization control utilizing a fuzzy neural network PID and conventional control methodologies. Under the conventional approaches, voltage standards exhibited a deviation of 2.5% along with a fluctuation span that reached as high as 5%. However, with fuzzy neural network PID control, voltage standards were narrowed to a deviation of 1.5%, with a fluctuation range reduced to 3%. Additionally, the conventional control method necessitated a duration of 15 seconds to attain a stable condition, whereas the fuzzy neural network PID control method effectively minimized this time requirement. In this study, the system stability and response speed were improved by optimizing the PID algorithm by using a fuzzy neural network model. Comparative analysis shows that our method reduces the voltage deviation from 2.5% to 1.5% and reduces the fluctuation range from 5% to 3%. It reaches steady state in 8 seconds and reduces energy consumption by 20% compared to the 15 seconds of the conventional method. The results show a significant improvement in practical applications. Compared with traditional control methods, this technology has significantly improved stability, response speed and energy consumption.

Author 1: Yabing Li

Author 2: Limin Su

Author 3: Huili Guo

Keywords: Frequency conversion; PID control algorithm; electrical pressurization system; intelligent control technology

Paper 103: An Open-Domain Search Quiz Engine Based on Transformer

Abstract: As the volume of information on the Internet continues to grow exponentially, efficient retrieval of relevant data has become a significant challenge. Traditional keyword matching techniques, while useful, often fall short in addressing the complex and varied queries users present. This paper introduces a novel approach to automated question and answer systems by integrating deep learning and natural language processing (NLP) technologies. Specifically, it combines the Transformer model with the HowNet knowledge base to enhance semantic understanding and contextual relevance of responses. The proposed system architecture includes layers for word embedding, Transformer encoding, attention mechanisms, and Bi-directional Long Short-Term Memory (Bi-LSTM) processing, enabling sophisticated semantic matching and implication recognition. Using the BQ Corpus dataset in the banking and finance domain, the system demonstrated substantial improvements in accuracy and F1-score over existing models. The primary contributions of this research are threefold: (1) the introduction of a semantic fusion approach using HowNet for enhanced contextual understanding, (2) the optimization of Transformer-based deep learning techniques for Q&A systems, and (3) a comprehensive evaluation using the BQ Corpus dataset, demonstrating significant improvements in accuracy and F1-score over baseline models. These contributions have important implications for improving the handling of complex and synonym-rich queries in automated Q&A systems. The experimental results highlight that the integrated approach significantly enhances the performance of automated Q&A systems, offering a more efficient and accurate means of information retrieval. This advancement is particularly crucial in the era of big data and Web 3.0, where the ability to quickly and accurately access relevant information is essential for both users and organizations.

Author 1: Xiaoling Niu

Author 2: Ge Guo

Keywords: Natural language processing; deep learning; transformer; Bi-LSTM; semantic understanding

Paper 104: An Artificial Neural Network Model for Water Quality Prediction in the Amoju Hydrographic Subbasin, Cajamarca-Peru

Abstract: Water quality is crucial for sustaining life, and accurate prediction models are essential for effective management. This study introduces an Artificial Neural Network (ANN) model designed to predict the Water Quality Index (WQI) in the Amoju Hydrographic Subbasin, Cajamarca-Peru. The model was developed using key water quality parameters, including electrical conductivity (EC), total dissolved solids (TDS), calcium carbonate (CaCO3), and phosphate (〖PO〗_4^(3-)), identified through Pearson correlation analysis. Data from water samples collected over six months were used to train and validate the model. Results revealed that the ANN model achieved high predictive accuracy, with a significant correlation between WQI and the aforementioned parameters. The model's performance outstrips traditional methods demonstrating its capability to effectively capture complex interdependencies among water quality indicators. This research emphasizes the potential of AI-driven approaches for enhancing predictive accuracy in environmental monitoring. Future studies should consider incorporating additional variables, such as heavy metals and microbial indicators, and consider the application of real-time AI-driven monitoring systems to further refine water quality management strategies. The ANN model presented here offers a promising tool for decision-makers, providing a reliable method for predicting water quality in similar hydrographic basins and contributing to the broader field of AI in environmental science.

Author 1: Alex Alfredo Huaman Llanos

Author 2: Jeimis Royler Yalta Meza

Author 3: Danicza Violeta Sanchez Cordova

Author 4: Juan Carlos Chasquero Martinez

Author 5: Lenin Quiñones Huatangari

Author 6: Dulcet Lorena Quinto Sanchez

Author 7: Roxana Rojas Segura

Author 8: Alfredo Lazaro Ludeña Gutierrez

Keywords: Artificial neural networks; hydrographic subbasin; machine learning models; water quality index; water resource management

Paper 105: Interactive ChatBot for PDF Content Conversation Using an LLM Language Model

Abstract: Natural Language Processing (NLP) leverages Artificial Intelligence (AI) to enable computer programs to understand and generate human language. ChatGPT has recently become popular in assignment accomplishment. This project aims to develop and improve an interactive PDF chat application using OpenAI's language model (LLM), specifically GPT-3.5, integrated with Streamlit and LangChain frameworks to assist in learning process. The application enhances user interaction with documents by providing real-time text extraction, summarization, translation, and user-defined question-answering to increase learning opportunities. Key features include obtaining document summaries, multilingual support for improved accessibility, and a document preview section with features such as zoom, rotation, and download. Although it currently faces limitations in handling image-rich PDFs, future enhancements include better image rendering, conversation history, and query download features. Overall, this interactive chatbot model aims to streamline document interaction, making information retrieval efficient and user-friendly.

Author 1: Ting Tin Tin

Author 2: Seow Yu Xuan

Author 3: Wong Man Ee

Author 4: Lee Kuok Tiung

Author 5: Ali Aitizaz

Keywords: Natural language processing; learning opportunities; ChatGPT; PDF

Paper 106: Prototype of an Indoor Pathfinding Application with Obstacle Detection for the Visually Impaired

Abstract: This study presents an initial prototype for a project aimed at assisting visually impaired individuals using deep learning techniques. The proposed system utilizes the You Only Look Once (YOLOv8) algorithm to detect objects tagged as obstacles. Designed for indoor environments, the system employs a CCTV camera and a computer server running the YOLOv8 model. Additionally, the A-star algorithm is used to determine the optimal path to avoid detected obstacles. Video frames are divided into tiles, each considered a node; nodes with detected objects are marked with a value of 0. The YOLOv8 model currently achieves an initial accuracy rate of 70%, with a mean Average Precision (mAP) at an Intersection over Union (IoU) threshold of 0.5 reaching 0.993 across all classes. This high mAP indicates an exceptional balance between precision and recall, signifying the model's effectiveness in object detection. Furthermore, the model yields an impressive F1-score of 0.99 at a confidence threshold of 0.624, demonstrating a robust balance between precision and recall, which is crucial for minimizing both false positives and false negatives. This prototype being developed assumes that a destination can be set by an operator of the system using the server that connects to the CCTV camera. The system was tested in enclosed environments and was able to provide a path that potentially avoids obstacles. The development of audio commands to guide visually impaired users is ongoing. These audio commands depend on identifying the direction an individual is going, requiring an additional deep-learning model to generate accurate instructions.

Author 1: Ken Gorro

Author 2: Lawrence Roble

Author 3: Mike Albert Magana

Author 4: Rey Paolo Buot

Author 5: Louis Severino Romano

Author 6: Herbert Cando

Author 7: Bonifacio Amper

Author 8: Rhyan Jay Signe

Author 9: Elmo Ranolo

Keywords: Yolov8; A-star algorithm; pathfinding; deep learning

Paper 107: Real-Time Road Damage Detection System on Deep Learning Based Image Analysis

Abstract: This research paper introduces a sophisticated deep learning-based system for real-time detection and segmentation of road damages, utilizing the Mask R-CNN framework to enhance road maintenance and safety. The primary objective was to develop a robust automated system capable of accurately identifying and classifying various types of road damages under diverse environmental conditions. The system employs advanced convolutional neural networks to process and analyze images captured from road surfaces, enabling precise localization and segmentation of damages such as cracks, potholes, and surface wear. Evaluation of the model's performance through metrics like accuracy, precision, recall, and F1-score demonstrated high effectiveness in real-world scenarios. The confusion matrix and loss curves presented in the study illustrate the system's ability to generalize well to unseen data, mitigating overfitting while maintaining high detection sensitivity. Challenges such as variable lighting, shadows, and background noise were addressed, highlighting the system's resilience and the need for further dataset diversification and integration of multimodal data sources. The potential improvements discussed include refining the convolutional network architecture and incorporating predictive maintenance capabilities. The system's application extends beyond mere detection, promising transformative impacts on urban planning and infrastructure management by integrating with smart city frameworks to facilitate real-time, predictive road maintenance. This research sets a benchmark for future developments in the field of automated road assessment, pointing towards a future where AI-driven technologies significantly enhance public safety and infrastructure efficiency.

Author 1: Bakhytzhan Kulambayev

Author 2: Belik Gleb

Author 3: Nazbek Katayev

Author 4: Islam Menglibay

Author 5: Zeinel Momynkulov

Keywords: Deep learning; road damage detection; Mask R-CNN; image segmentation; convolutional neural networks; infrastructure management; smart cities; real-time analytics; predictive maintenance; urban planning

Paper 108: Real-Time Sign Language Fingerspelling Recognition System Using 2D Deep CNN with Two-Stream Feature Extraction Approach

Abstract: This research paper introduces a novel sign language recognition system developed using advanced deep learning (DL) techniques aimed at enhancing communication capabilities between deaf and hearing individuals. The system leverages a convolutional neural network (CNN) architecture, optimized for the real-time interpretation of dynamic hand gestures that constitute sign language. A comprehensive dataset was employed to train and validate the model, encompassing a diverse range of gestures across different environmental settings. Comparative analysis revealed that the deep learning-based model significantly outperforms traditional machine learning techniques in terms of recognition accuracy, particularly with the increase in the volume of training data. This was illustrated through various performance metrics, including a detailed confusion matrix and Levenshtein distance measurements, highlighting the system’s efficacy in accurately identifying complex gestures. Real-time application tests further demonstrated the model's robustness and adaptability to varying lighting conditions and backgrounds, essential for practical deployment. Key challenges identified include the need for broader linguistic diversity in training datasets and enhanced model sensitivity to subtle gestural distinctions. The paper concludes with suggestions for future research directions, emphasizing algorithm optimization, data diversification, and user-centric design improvements to foster wider adoption and usability. This study underscores the potential of deep learning technologies to revolutionize assistive communication tools, making them more accessible and effective for the deaf community.

Author 1: Aziza Zhidebayeva

Author 2: Gulira Nurmukhanbetova

Author 3: Sapargali Aldeshov

Author 4: Kamshat Zhamalova

Author 5: Satmyrza Mamikov

Author 6: Nursaule Torebay

Keywords: Deep learning; sign language recognition; convolutional neural networks; real-time processing; gesture recognition; machine learning; accessibility technology