http://repository.mut.ac.ke:8080/xmlui/handle/123456789/43 Mon, 20 Apr 2026 10:45:25 GMT 2026-04-20T10:45:25Z http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6815 A HETEROGENEOUS DEEP ENSEMBLE APPROACH FOR ANOMALY DETECTION IN CLASS IMBALANCED ENERGY CONSUMPTION DATA Marangu, David Kaimenyi; Wanjau, Stephen Kahara The integrity and efficiency of modern energy grids are increasingly reliant on accurate anomaly detection within energy consumption data. However, class imbalance poses significant challenges, where normal consumption patterns vastly outnumber critical anomalies, leading to biased detection models. This paper presents a novel heterogeneous deep ensemble model specifically designed to handle class imbalance in energy consumption anomaly detection. The architecture strategically integrates Bidirectional Long ShortTerm Memory (BiLSTM) networks for capturing temporal dependencies and Convolutional Neural Networks (CNNs) for feature extraction. Cost-sensitive learning was incorporated to address class imbalance, with rigorous hyperparameter tuning using Bayesian optimization. The model was evaluated using the State Grid Corporation of China (SGCC) dataset containing 42,372 customers' electricity consumption data.The deep ensemble model achieved impressive performance metrics: accuracy of 97.5%, precision of 97%, recall of 99%, F1-score of 98%, and AUC-ROC score of 99%. Statistical analysis confirmed significant improvements over baseline methods (BiLSTM and CNN) and existing ensemble models, with p-values consistently below 0.05.The heterogeneous ensemble architecture demonstrates superior performance compared to individual models and existing approaches. Cost-sensitive learning effectively addresses class imbalance while maintaining high accuracy. The findings establish new performance benchmarks for anomaly detection in energy systems with significant implications for energy efficiency, grid stability, and infrastructure security. Wed, 01 Jan 2025 00:00:00 GMT http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6815 2025-01-01T00:00:00Z http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6807 Natural Language Processing with Transformer-Based Models: A Meta-Analysis Munyao, Charles The natural language processing (NLP) domain has witnessed significant advancements with the emergence of transformer-based models, which have reshaped the text understanding and generation landscape. While their capabilities are well recognized, there remains a limited systematic synthesis of how these models perform across tasks, scale efficiently, adapt to domains, and address ethical challenges. Therefore, the aim of this paper was to analyze the performance of transformer-based models across various NLP tasks, their scalability, domain adaptation, and the ethical implications of such models. This meta-analysis paper synthesizes findings from 25 peer-reviewed studies on NLP transformer-based models, adhering to the PRISMA framework. Relevant papers were sourced from electronic databases, including IEEE Xplore, Springer, ACM Digital Library, Elsevier, PubMed, and Google Scholar. The findings highlight the superior performance of transformers over conventional approaches, attributed to selfattention mechanisms and pre-trained language representations. Despite these advantages, challenges such as high computational costs, data bias, and hallucination persist. The study provides new perspectives by underscoring the necessity for future research to optimize transformer architectures for efficiency, address ethical AI concerns, and enhance generalization across languages. This paper contributes valuable insights into the current trends, limitations, and potential improvements in transformer-based models for NLP Wed, 01 Jan 2025 00:00:00 GMT http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6807 2025-01-01T00:00:00Z http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6805 A Unified U-Net-Vision Mamba Model with Hierarchical Bottleneck Attention for Detection of Tomato Leaf Diseases Mutiso, Geoffry; Ndia, John Tomato leaf diseases significantly reduce crop yield; therefore, early and accurate disease detection is required. Traditional detection methods are laborious and error-prone, particularly in large-scale farms, whereas existing hybrid deep learning models often face computational inefficiencies and poor generalization over diverse environmental and disease conditions. This study presents a unified U-Net-Vision Mamba Model with Hierarchical Bottleneck Attention Mechanism (U-net-Vim-HBAM), which integrates U-Net’s high-resolution segmentation, Vision Mamba’s efficient contextual processing, and a Hierarchical Bottleneck Attention Mechanism to address the challenges of disease detection accuracy, computational complexity, and efficiency in existing models. The model was trained on the Tomato Leaves and PlantVillage combined datasets from Kaggle and achieved 98.63% accuracy, 98.24% precision, 96.41% recall, and 97.31% F1 score, outperforming baseline models. Simulation tests demonstrated the model’s compatibility across devices with computational efficacy, ensuring its potential for integration into real-time mobile agricultural applications. The model’s adaptability to diverse datasets and conditions suggests that it is a versatile and high-precision instrument for disease management in agriculture, supporting sustainable agricultural practices. This offers a promising solution for crop health management and contributes to food security. Wed, 01 Jan 2025 00:00:00 GMT http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6805 2025-01-01T00:00:00Z http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6769 SYSTEMATIC REVIEW OF VEHICULAR AD-HOC NETWORKS TRUST-BASED MODELS Nyutu, Samson Waweru; Ndia, John Gichuki; Mwangi, Peter Maina Vehicular Ad-hoc Networks (VANETs) are specialized type of Mobile Ad-hoc Networks (MANETs) developed to support vehicle-to-vehicle and vehicle-to-infrastructure communication. Security threats in these networks include malicious nodes, Man-in-the-Middle (MitM) attacks and Denial of Service (DoS) attacks which threaten network reliability. Trust-based models in VANETS seek to evaluate trust of nodes and data sharedand incorporate certain features which enhance adaptivity. This systematic literature review (SLR) analyses the adaptive trust-based models available in VANETs with an emphasis on the features that facilitate adaptability in the dynamic environment. The review follows Barbara Kitchenham (2007) systemic approach and accesses databases including Google Scholar, IEEE Xplore and Wiley Online Library where 34 articles are included. The findings highlight important characteristics such as historic data analysis, real-time behavior monitoring, adaptive trust score update, frequency of messages, context awareness and real-time detection of intrusive attacks that make the VANETs trust models more adaptable Wed, 01 Jan 2025 00:00:00 GMT http://repository.mut.ac.ke:8080/xmlui/handle/123456789/6769 2025-01-01T00:00:00Z