Disruptions in distributed systems can cause widespread failures and downtime which costs the company and lowers productivity. A recovery and fault tolerance system is essential because distributed system's complexity, unpredictability, and inner workings exacerbate failures. Predictive analytics has emerged in fault management, helping firms to move from reactive to proactive fault management. In this research, we examine the achievements of intelligent fault prediction approach that employs Gated Recurrent Units (GRU), Temporal Convolutional Networks (TCN), and Long Short-Term Memory (LSTM) networks to improve fault tolerance in distributed systems. The GRU and LSTM models can describe temporal, sequential data for a distributed system and identify and comprehend data changes. By employing these designs and TCNs, companies may better recognize fault patterns that emerge over time, forecast future failures, and increase fault management efficiency. TCNs identify long-range time dependencies and allow parallel processing to swiftly discover and respond to defects in largescale settings using causal and dilated convolutions. This framework uses deep learning models to examine system log and resource consumption data to identify probable failure symptoms and accurately predict future problems. The experimental results show that combining GRU, LSTM, and TCN models improves fault prediction accuracy and reduces unexpected downtime by identifying faults quickly and taking preventative action. Additionally, since the framework continuously collects data from the distributed system and monitors the logged information in real time, users can make better decisions and implement proactive responses to failures, proving that predictive analytics driven by deep learning technology increases intelligent fault tolerance in distributed systems. With 95% accuracy, 93% precision, 95% recall, 94% F1-score, and 0.97 ROC-AUC, the GRU + LSTM + TCN model outperforms all single and dual-model configurations. With a +7% accuracy boost over GRU, +5% over LSTM, and +4% over TCN, multi-model temporal feature fusion is beneficial for fault prediction.
The ever-increasing growth rate of the Internet of Things (IoT) has significantly increased network heterogeneity and data volumes, which have posed a significant risk of various cyber threats to resource-constrained IoT devices. Traditional Intrusion Detection Systems (IDS) may not perform well in such environments, which may be characterized by noisy data, class imbalance, and a large number of feature redundancies. This study proposes an Adaptive Spatiotemporal Defense Framework suitable for real-world IoT networks. The framework includes an efficient nonlinear data preprocessing approach, discriminative feature optimization, and a spatiotemporal learning framework to improve the performance of intrusion detection systems. The framework efficiently addresses the class imbalance problem, reduces redundant features, and maintains discriminative features, thereby improving the quality of the features and reducing the complexity. The inclusion of spatiotemporal learning in the framework enables the detection of complex data relationships with high precision. The performance of the proposed approach is evaluated on three benchmarking datasets: UNSW-NB15, ToN-IoT, and IoT-23. The accuracy obtained on these datasets is 89.61%, 97.94%, and 99.99%, respectively. Along with this, high F1-score results are obtained. This proves the enhancement in intrusion detection ability. Overall, the proposed framework proves to be a reliable, efficient, and lightweight solution for intrusion detection in heterogeneous IoT environments.
Orbital Angular Momentum (OAM) multiplexing enables high spectral efficiency but suffers from severe signal degradation due to beam divergence. Traditional compensation methods are computationally intensive and slow, making them unsuitable for real-time use. This paper proposes a machine learning-based solution using a Deep Neural Network (DNN) that learns to predict optimal beamforming weights directly from environmental factors like distance and turbulence. Trained on diverse channel data, the model achieves 96.2% of the optimal received power while reducing latency from 100 ms to just 0.5 ms--a 200 & times; improvement. It generalizes well to unseen scenarios, including mobile users and hardware impairments, outperforming conventional and adaptive baselines. Hardware-aware simulations confirm less than 5% performance degradation under realistic impairments, establishing a clear path to experimental validation. This work demonstrates a viable path toward adaptive, low-latency OAM systems for future 6G networks.
The Semantic Web extends the capabilities of the traditional Web by enabling machines to process and interpret data through ontology knowledgebase. Integrating ontologies into the Web facilitates more accurate and precise searches, task automation, and optimized integration between systems. This research work focuses on semantic information retrieval (SIR) for COVID-19-related queries, leveraging ontologies to generate precise search results and antonyms to reduce irrelevant results. By conducting syntactic and semantic analysis, the system expands the search query using the context derived from the ontology. The query is further refined by extracting antonyms via the ontology relations. The refined query is then submitted to the search engine to retrieve more precise results. A ranking module further filters and prioritizes the most pertinent result links. The SIR approach is novel among existing information retrieval systems in that it eliminates irrelevant search results via antonyms, rather than displaying all the retrieved results based on the query, and in that it re-ranks the results semantically. The SIR algorithm demonstrates significant performance improvements for most queries, primarily due to the semantic analysis, antonyms addition and re-ranking processes. The query dataset achieved 100% precision and 80% recall, outperforming existing search engines in these metrics.
Legal Judgment Prediction (LJP) has recently gained significant interest from both academic and legal practitioners. The majority of LJP methods focus on single label prediction problem, neglecting the real-world multilabel case. Therefore, this study aimed to classify multilabel legal cases using label correlation and problem transformation methods. Data were collected from a publicly accessible legal document in the European Court of Human Rights (ECHR) and EUR-Lex. Multilabel text classification tasks face challenges such as sample diversity, complexity, and the need for effective utilization of label correlations. In this paper, we propose a model that integrates domain specific text embedding and label correlation. Proposed model leverages label powerset as problem transformation to transform a multilabel problem to a multiclass problem by incorporating domain specific text embedding and label correlation, which enhances classification performance in charge prediction and addresses label omission issues. Extensive experiments on two legal text datasets demonstrate the model's excellent performance. The proposed model substantially outperformed two baseline studies by attaining competitive results of 80.32%-90.09% F1-score and 0.0119-0.0210 Hamming Loss score, respectively. Meanwhile, the baseline models have attained 52%-80% F1-score and 0.0452-0.1479 Hamming Loss score. Proposed model's performance significantly surpasses the baseline models. The significance of this study is the implementation of label correlation in label powerset problem transformation method and the application of domain specific embedding to solve multilabel classification problem in legal domain.
Fake news detection is a highly crucial challenge in Natural Language Processing (NLP), particularly during significant social events like elections and national crises. This study uses the GPT-3.5-Turbo model to test the effectiveness of zero-shot and two-shot prompting in detecting fake news on the PolitiFact and Liar datasets. Zeroshot prompting consists of task instructions without examples, whereas two-shot prompting contains a few task-related examples. The methodology includes dataset preparation, Large Language Models (LLMs) response collection, encoding, and evaluation using metrics such as accuracy, precision, recall, and F1-score. The results show that twoshot prompting increases performance marginally across all parameters when compared to zero-shot prompting. PolitiFact's accuracy improved from 0.286 to 0.293, while Liar's improved from 0.220 to 0.226. Precision, recall, and F1-score also showed minor gains. However, these advances were not statistically significant and highlight the model's difficulty with handling multi-class classification in the political domain. The GPT-3.5-Turbo model performed better on the PolitiFact dataset, suggesting variability in performance across different datasets. In conclusion, although two-shot prompting provides a slight advantage, the GPT-3.5-Turbo's overall performance remains limited, indicating the need for more sophisticated techniques (such as advanced prompting methods or more powerful LLMs) to enhance fake news detection.
This study introduces an innovative approach to content-based image retrieval (CBIR) specifically designed for fish species identification. The proposed method integrates shape, colour, and texture features using Zernike Moments Invariant (ZMI) and Local Directional Pattern (LDP), applied to the momentgram and the hue channel of the HSV colour space. This fusion ensures invariance to transformations such as rotation, scaling, and translation, enabling robust performance on natural images with varying orientations and quality. The method was evaluated using the Fish4Knowledge dataset, consisting of 27,370 images, with 30% randomly selected as query images. Experimental results demonstrate that the proposed method achieved a mean average precision (MAP) of 84.17%, significantly outperforming comparable state-of-the-art approaches. Statistical analysis using two-tailed paired t-tests confirms its superiority. By combining global shape descriptors, local texture features, and colour properties, this method delivers a comprehensive representation of fish images. The inclusion of moment-based descriptors enhances its robustness against low-resolution images and noise. This research underscores the importance of combining diverse features within CBIR systems and offers a significant improvement in retrieval accuracy, contributing to domain-specific applications such as sustainable fisheries management and aquaculture research.
Noisy text normalization is a critical preprocessing step in natural language processing (NLP), particularly for user-generated content (UGC) that contains a lot of slang, abbreviations, and typographical errors. This extended study investigates the performance of multiple similarity measures in normalizing Malay noisy text, addressing gaps in prior study that predominantly relied on rule-based approaches and single similarity measures. By systematically evaluating token-based, edit-based, and sequence-based similarity measures across various thresholds, this study provides a comprehensive analysis of their effectiveness and computational efficiency. The methodology comprises a two-phase experiment: an initial phase to identify optimal thresholds using a small dataset and a second phase that generalizes findings on a larger dataset. Key findings reveal that edit-based measures, such as Levenshtein Distance and Damerau-Levenshtein, consistently outperform other measures at lower thresholds, achieving normalization success rates exceeding 83%. Ratcliff/Obershelp emerged as the most effective sequence-based measure, while token-based measures like Jaccard and Cosine demonstrated limited performance. The study also highlights the critical role of threshold in balancing normalization accuracy and flexibility. Additionally, computational time analysis underscores the trade-offs between accuracy and efficiency across similarity categories. These findings pave the way for more robust and adaptable text normalization strategies, particularly for Malay language studies.
Deep Reinforcement Learning (DRL), a subfield of machine learning, has shown remarkable potential in various domains, including recommender systems (RSs). This study leverages DRL to improve RS performance by effectively modeling user preferences and addressing their unique needs. A knowledge graph (KG) is constructed using product information, such as features and historical purchase data, to serve as the environment for the Markov Decision Process (MDP) within the DRL framework. The KG is enriched with embeddings to enable efficient navigation and enhance its utility. The Actor-Critic model in DRL employs these embeddings within the MDP, enabling a more accurate representation of user preferences. Central to this approach is the Representation of User Preferences via Path Embedding Propagation (RUPPEP), which serves as the study's core contribution. Experimental results demonstrate that DRL-based RSs achieve superior performance metrics, with a 13.26% improvement in NDCG for the Amazon Cell Phones dataset and a 15.43% increase for the Amazon Beauty dataset compared to the best SOTA baseline model, highlighting their potential to advance the field of recommendation systems.
Calculating impact factors solely from raw citation counts can be misleading because citation counts vary across disciplines and publication years. Citation analysis is a fundamental bibliometric methodology that aids in determining trends, frequency, and influence; nevertheless, normalisation methods for reducing skewness are still not well understood and are not always used. Selecting an appropriate normalisation technique to map source data to a standardised citation scale effectively is challenging. This study aims to provide new insights into which normalisation methods researchers should use when conducting citation analysis. Normalisation at the author level using Field-Weighted Citation Impact (FWCI) and Log Normalisation Citation Score (Log NCS), and the field level using Mean Normalised Citation Score (MNCS) and Mean Normalised Log Citation Score (MNLCS) are the two components of the comparative analysis presented in this work. Using citation data from six different academic fields, this study assesses how well each approach reduces skewness and creates more equitable comparisons. This study offers four contributions: it provides a structured comparison approach to author and field normalisation; it empirically demonstrates conditions under which log transformation outperforms non-log methods; it provides decision-making guidance for researchers when selecting normalisation approaches; and it validates findings across multiple disciplines. These findings are intended to improve the accuracy and validity of citation-based impact evaluations, thereby facilitating more equal academic benchmarking and collaboration.
Creating an automated pain detection system for infants less than a year old is essential because they are unable to communicate their discomfort verbally. Conventional assessment techniques like FLACC (Face, Legs, Activity, Cry, Consolability) require considerable time and may not be effective for infants with vocal cord impairments. Utilizing infants' facial expressions for real-time, automated pain detection presents a promising approach that facilitates rapid medical response. This study adopts a machine learning approach using infant facial expressions as input and explores the efficacy of various MobileNetV3 architectures, both Small and Large, enhanced with attention mechanisms. We introduced modifications involving 12 model variants, including the integration of CBAM (Convolutional Block Attention Module), ECA (Efficient Channel Attention), and SAM (Spatial Attention Module) attention modules, as well as hybrid attention configurations (ECA + CBAM and ECA + SAM). Training was conducted on a FLACC-based dataset comprising 56 videos collected from infants under 12 months undergoing hernia treatment at Dr. Soetomo General Hospital, Surabaya, East Java, Indonesia, from November 2011 to December 2022. The dataset is categorized into three pain levels: no pain, low/moderate pain, and severe pain. Results demonstrate that attention mechanisms significantly enhance model accuracy, with hybrid configurations consistently achieving the best performance. The ECA + CBAM hybrid configuration achieved the highest accuracy of 94.5%, representing a 5% improvement over baseline models, while also reducing misclassifications across all pain levels. However, these gains come with increased computational complexity, including higher parameter counts, greater FLOPs, longer inference times, and higher memory usage. These results indicating their robustness in real-time pain detection for infants, thereby highlighting their potential for practical clinical applications.
Existing access control provides a security solution to manage BYOD policies but is limited to controlling and providing adequate security. This paper comprehensively implements access control encompassing three security layers of the BYOD policy simultaneously: tactical, strategic, and operational. This system comprises the initial component and dynamic attributes for enforced access decisions. The second component consists of risk monitoring and anomaly detection algorithms. Finally, the third component employs the adaptive policy adjustment algorithm, which provides recommendations to the administration for policy updates in cases of abnormal access based on the results of the attack detection algorithm. The suggested access control solution was implemented using machine learning algorithms to detect anomalous and atypical user behavior. The experimental results obtainedfrom the UNSW-NB15 dataset confirmed that the proposed access control could improve the anomaly detection algorithm and adaptive policy adjustment performance while reducing prediction detection time. The results demonstrated that the risk monitoring and anomaly detection algorithm, with a prediction time of 0.5 seconds and an accuracy rate of 0.95 percent, is the most effective method for monitoring attacks. Additionally, the results indicated that the accuracy of the adaptive policy adjustment algorithm was approximately 97%, with a threshold value of 0.26 being the optimal modification threshold value. The solution could enhance the detection of insider threats, access control, and policy management while at the same time making access control dynamic, adaptable, flexible, and secure.
Flooding poses serious risks to lives, infrastructure, and ecosystems, underscoring the need for accurate forecasting. However, missing values in hydrological datasets-often caused by equipment failure or extreme weather-can compromise forecast reliability. This study evaluates five imputation techniques: Last Observation Carried Forward, Next Observation Carried Backward, Linear Interpolation, Spline Interpolation, and K-Nearest Neighbours, to identify the most effective method for reconstructing missing flood-related data. Using temperature, humidity, and water level records from the Langat River, Selangor, Malaysia, each method's performance was assessed via Root Mean Square Error. Results show that Linear Interpolation generally yields the lowest error, while Next Observation Carried Backward performs best when missing data is minimal (1.20%).
This study addresses the challenges posed by imbalanced data in multilevel depression classification by leveraging the Adaptive Synthetic (ADASYN) technique. Subject Matter Experts (SMEs) annotate data collected from X into four categories: None, Mild, Moderate, and Severe. The imbalanced distribution, particularly with a larger group for the None category, prompts the application of ADASYN for effective data augmentation. The research framework encompasses Data Collection, Expert Data Annotation, Text Preprocessing, and Text Representation and Classification. Evaluation metrics, including Recall and F1 score, gauge the model's effectiveness in multilevel depression classification. Results showcase the efficacy of the ADASYN-enhanced model, specifically with XGBoost, demonstrating improved classification accuracy, especially for minority classes. This study contributes valuable insights to the field of multilevel depression classification, emphasizing the effectiveness of ADASYN in managing imbalanced data scenarios and showcasing the applicability of XGBoost in enhancing model performance.
Generative Artificial Intelligence (GAI) has rapidly disseminated within a brief period. Given GAI's early development and widespread presence, significant issues have arisen concerning the training and functioning of its underlying models, especially in ethical domains such as those related to ChatGPT and NLP. Thus, it is imperative to engage in more extensive discourse on the subject. Despite these advancements, GAI continues to pose substantial ethical concerns, such as the potential for the fabrication of scientific results which creates false information. These concerns underscore the necessity of addressing ethical challenges to maintain the integrity and efficacy of educational and research environments. This study aims to critically analyze the ethical challenges, limitations, and potential studies for adopting GAI systems through a systematic literature review. We utilized the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology to examine and evaluate relevant research studies. The research targeted articles published in 2023 and afterward. This systematic search was conducted from May 11 to May 17, 2024. From 712 records retrieved from nine academic databases, 9 duplicate record papers, 503 publications were determined to be incomplete and unrelated, 46 papers cannot be retrieved, and 28 are not peer-reviewed. As a result, twenty-three (23) research publications were found from 126 papers were qualified for consideration. The systematic literature review revealed that challenges related to the existing ethical framework for GAI adoption include addressing ethical concerns, establishing an evaluation model, formulating global principles for the ethical use and development of GAI-based systems, creating frameworks to regulate the ethical and responsible use of GAI, and addressing privacy and security issues, alongside the necessity for clear guidelines and ethical parameters for ethical GAI. The limitations identified in recent studies include insufficient empirical evidence and validation, a need for practical implementation, insufficient specific guidelines, insufficient evaluation metrics, and measurement instruments, and the necessity for further exploration of ethical considerations. The prospective studies within the current research framework involve the development of practical implementation standards, guidelines, and best practices; the creation of evaluation metrics; the formulation of regulatory frameworks to ensure ethical use; the collection of stakeholder perspectives; and the exploration of ethical implications alongside industry-specific analyses. Overall, this review can be used as a guide for researchers and all interested parties to encourage further research and experimentation related to ethical GAI adoption in the future.
The integration of deep learning technologies into agriculture has the potential to revolutionize smart farming by enhancing efficiency, sustainability, and productivity. This study focuses on leveraging the Hippopotamus Optimized Attention Hierarchically Gated Recurrent Algorithm (HOA-HGRA) within a containerized environment to analyze and predict critical agricultural variables such as weather patterns, crop yield, and soil moisture. The proposed methodology involves containerizing deep learning models like HOA-HGRA and orchestrating them with Kubernetes on HPC clusters. This enables precise monitoring and management of crop growth, soil conditions, and livestock health, ensuring optimal resource utilization and enhanced productivity. The hyperparameters tuning and the performance optimization are performed by applying the Oppositional Hippopotamus optimization with opposition learning-based strategy. The overall performance of the AHGR-OH model is validated by utilizing the France-CGIAR BRIDGE, SmartAgriculture, Smart precision agriculture, Smart Farming Irrigation Systems, and IoT in Smart Farming Market Report datasets. Moreover, key metrics such as latency, precision, F1-score, recall, scalability, accuracy, MSE, and ROC are utilized to estimate the effectiveness of the AHGR-OH method. By comparing, the developed method grants 2s latency, 0.5 MSE, higher scalability, precision, F1-score, accuracy, and recall of 98.5%, 97.9%, 97.4%, 99.1%, and 97.9% respectively. This paper demonstrates the potential of the AHGR-OH Algorithm to revolutionize smart farming practices.
As e-commerce transactions continue to surge, the threat of fraud has escalated, posing significant challenges due to class imbalances, rapidly evolving fraud tactics, and the critical need to balance false positives and negatives. This study effectively addresses these challenges through an advanced ensemble stacking approach, integrating Support Vector Machine (SVM), Neural Network, Gradient Boosting, and AdaBoost as base models, with a Random Forest as meta-model to deliver final predictions. Using an e-commerce transaction dataset, our approach achieved 99.87% accuracy, significantly outperforming individual models. The meta-model further demonstrated 0.99 precision, 0.98 recall, and 0.99 F1-score for fraud cases (Class 1), highlighting its strong ability to accurately detect fraudulent transactions while minimizing false positives and false negatives. While SVM had the longest execution time, the Neural Network was the most efficient, and AdaBoost contributed the most to the meta-model's predictions. Model validation was performed using Local Interpretable Model-Agnostic Explanations (LIME), highlighting Transaction Hour, Transaction Amount, and Account Age Days as key predictive features. The model was successfully deployed to a web-based application, demonstrating real-time fraud detection capabilities. This research offers a robust, interpretable method for e-commerce fraud prevention, potentially reducing financial losses and enhancing online transactions.
Floods are recurrent natural disasters that can have a devastating impact on societies, economies, and the environment. Therefore, it is critical to predict and manage flood situations promptly to minimize the damage they cause. However, many people are unaware of flood risks, and there are limited mobile applications that can provide timely and accurate flood predictions. This study explores the application of the Support Vector Machine (SVM) algorithm for flood prediction in a mobile application. The aim is to provide users with timely and accurate flood predictions, enabling them to make informed decisions and take necessary precautions to mitigate flood impacts. By integrating the SVM algorithm into a mobile application, users gain convenient access to flood predictions, empowering them to be better prepared for potential flooding events. The user-friendly platform delivers critical flood forecasts, ensuring individuals and communities can respond effectively to flood situations. The evaluation of the SVM algorithm's performance reveals an achieved accuracy of 66.66%. In conclusion, this study underscores the potential of the SVM algorithm for flood prediction in a mobile application. These findings contribute to the field of flood forecasting technology, paving the way for more sophisticated and effective flood prediction tools in the future.
The number of cyberattacks caused by malware targeting critical sectors, such as energy systems, telecommunications, healthcare, and finance, is rapidly increasing worldwide. The evolution of malware has made detection techniques more challenging, resulting in financial losses and reduced productivity. To address this issue, this paper presents the Enhancement of Parallel Deep Learning for Malware Detection (EPDL-MD) model. This model focuses on improving the parallel convolutional neural network (CNN) architecture. The performance of the CNN is influenced by its hyperparameters, and the enhancements have led to an increase in accuracy and learning rate. The experiment utilized 176,000 malware samples, which were sourced from 86 distinct malware families and one benign family. Based on the analysis and experiments, the EPDL-MD model has achieved an impressive accuracy rate of 99%.
In distributed systems with failure interruption, the performance of database replication transactions might become very critical. Any distributed system that enforces data replication can be impacted by this problem. The fault tolerance approach is crucial to ensure the data replication transactions are always effective and dependable despite failures. The key advantage of fault tolerance is its capacity to complete the transaction notwithstanding a failure and restore system availability. This paper proposes a fault tolerance approach namely Binary-Vote-Assignment-Grid with High-Priority-Neighbour (BVAGHPN). It improves the efficiency of the data replication transaction in term of total execution time. This approach combines BVAG data replication transaction manager with the HPN to manage the transaction in the event of disasters. Instead of waiting for the problem to be fixed in the event of disaster, BVAGHPN halts the transaction on a failure replica, remove the failing replica from the alive quorum, and proceed the transaction with other replicas based on its own rating. BVAGHPN improves the outcomes of BVAG and BVAGCR in terms of the total execution time for two cases, PR failure and NR failure. For PR failure, BVAGHPN exceeds BVAG with 69.02% and BVAGCR (54.67%), respectively. Meanwhile, for NR failure, BVAGHPN improves BVAG with 76.88% and BVAGCR (71.97%).
