Botswana International University of Science and Technology
BIUST Institutional Repository
Discover and explore the research, scholarship and creative works of the BIUST community.
Browse Repository
Explore BIUST research through structured browse categories.
Explore BIUST Communities
Browse the repository's main research and content categories.
BIUST Webinar Series
This community is made up of webinar series presented by BIUST staff and students.
Explore community →Books, Chapters and Sections
This community is made up of electronic copies of books, chapters and sections produced by staff and students of Botswana International University of Science and Technology.
Explore community →Conference/Symposium/Workshop/Seminar Works
The community is made up of conference proceedings, papers and presentations from conferences, symposiums, workshops and events held at or sponsored by Botswana International University of Science and Technology( BIUST).
Explore community →Electronic Theses and Dissertations
The community is made up of electronic theses and dissertations produced by post graduate students of Botswana International University of Science and Technology( BIUST)
Explore community →Research Articles
This collection is made up of research articles produced by both staff and post graduate students of BIUST community.
Explore community →Speeches/Lecturer Series
This community is made up of speeches and lectures series produced by BIUST staff and students.
Explore community →Latest Submissions
Discover the latest materials added to the BIUST Institutional Repository.
Item type:Item, An ontology-driven framework for animal traceability(Botswana International University of Science and Technology (BIUST), 2025-07) Mokgetse, Tshepiso Larona; Hlomani, Hlomani; Sigwele, TshiamoTraceability systems are crucial for regulatory compliance and consumer trust in food supply integrity and animal protection. Animal traceability systems ensure food safety and disease outbreak control. However, the current systems lack interoperability and are badly integrated across multiple stakeholders. The non-interoperable animal traceability systems pose challenges for farmers and the government in tracking stray animals, diseases, feeding, vaccinations, and livestock export regulations. The main cattle identification and traceability system being used in Botswana is the non-interoperable Botswana Livestock Identification and Traceability System. The six stakeholders concentrate on cow but lack a single system for interoperability, integration, standardization, and domain understanding among them. This research proposes the development of an interoperable and ontology-driven animal traceability framework to solve the gap of poor interoperability in current frameworks for animal traceability, with Botswana as a case study. Limitations of current traceability frameworks were identified, and the components for the design and development of the proposed interoperability framework were generated using the mixed method design methodology - qualitative and quantitative research. Involving robust data analysis. Quantitative identified system gaps and patterns in traceability practices though descriptive statistics. Qualitative research carried out a thematic analysis using NVivo tool on stakeholder interviews and documents, for in-depth understanding of socio-technical aspects of traceability systems. Together the designs reinforced the validity and reliability of the findings. The analysis identified cattle, farmers, system, and records as the key concepts. 39% of Botswana participants find BAITS ineffective, making it the main challenge of animal traceability. An interoperable ontology-driven animal traceability framework is developed. In addition, a new and novel animal traceability ontology was developed using the METHONTOLOGY ontology development methodology, covering knowledge acquisition, specification, conceptualization, formalization, and evaluation. The process ensures a transparent and systematic ontology development of the animal traceability domain to model the key traceability components. The ontology was evaluated for quality based on consistency, clarity, adaptability, and coverage. The Ontometric evaluation tool was used to evaluate revealing a high level of inheritance richness 0.692308, relationship richness 0.678571, and average depth 0.9, indicating detailed knowledge ontology, strong class connections, and seamless data integration and exchange. The evaluation of reference and candidate ontologies revealed 66.67% of animal traceability ontology is more interoperable than animal health ontology at 33.33%. Altogether, results indicate that the framework achieved interoperability and standardization, producing seamless integrated systems. Overall, this study contributes both as a template that is methodological for agricultural research in semantic systems. Also, providing a foundation for adaptable solutions to support worldwide supply chain compliance and strategies in animal health. The ontology-driven framework is designed as a guiding principle for system developers, a blueprint to implement semantically correct integrated traceability. The ontology is developed to contribute semantic interoperability and can be integrated with data infrastructures already existing.Item type:Item, Design of iterative convolutional neural network model for malicious node detection and optimization in wireless sensor networks(Botswana International University of Science and Technology, 2025-01) Moka, Moemedi; Selvaraj, Rajalakshmi; Kuthadi, VenumadhavWireless Sensor Networks (WSNs) represent a critical advancement in modern communication technologies, characterized by distributed sensor nodes that communicate wirelessly to monitor and report environmental conditions. These networks are integral to applications ranging from military surveillance and environmental monitoring to smart grids and intelligent transportation systems. The reliability and security of WSNs are paramount, as they operate in dynamic and often hostile environments where nodes can be susceptible to various attacks, including malicious node injections. Therefore, this research develops a novel model—an Iterative Convolutional Neural Network-based Malicious Node Detection (ICNNMND)—that provides an innovative three-phase method to enhance security and performance in WSNs. In the First phase of work, it collects the sample node dataset and performs pre-processing with malicious node detection utilising machine learning techniques. A novel algorithm called Advanced Dataset Pre-processing Algorithm (ADPA) was developed for pre-processing of nodes and Malicious Node Detection with Machine Learning Secure Learning (MND-MLSL) algorithm for handling the data preparation, attributes of nodes, topologies to be followed and the communication packets to be normalised for better transmission. The machine learning models were utilised based on the neural network schema and the experiments were completed. The results in first phase showed that the packets delivery of 21 instances out of 100 were found to be stolen and delayed with the execution time of 185.5513 seconds, stolen rate of 2% and total energy consumption of 562.7417 joules respectively. The packet delivery rate was 98% in the absence of malicious nodes. In the second phase, Iterative Convolutional Neural Network based Malicious Node Detection (ICNNMND) was designed which identifies, the complex patterns during data transmission to perform classification and also maximises pool layers and fully connected layers to enhance robustness using dropout regularisation technique. The accuracy was found to be 97.3155% after 100 iterations with good results in precision, recall and F1-Score respectively. The final third phase of work deployed the optimisation of ICNNMND using novel Network Hybrid Optimization Algorithm (NHOA) that enhances latency, reception quality and throughput for better optimisation. The results showed improvements in the overall latency and other significant parameters like Bit Error Rate (BER) getting reduced from 10-4 to 10-9 over 30 runs, indicating enhanced transmission accuracy and reduced errors. Throughput analysis also showed an enhanced output of 98% after 50 epochs, highlighting the model's effectiveness in maximizing data transfer efficiency. The model presented in the research showed that by combining machine learning with different optimisation and deep learning models, the security of data transmission can be enhanced with the high performance of WSNs.Item type:Item, Advancing bert pruning techniques: a conceptual framework and an ensemble method of neural subnetworks(Botswana International University of Science and Technology, 2025-02) Nyalalani, Smarts; Selvaraj, Rajalakshmi; Kuthadi, VenumadhavLarge Language Models (LLMs), such as BERT (Bidirectional Encoder Representations from Transformers), have experienced rapid growth and increasing complexity, which has introduced significant challenges related to computational costs, resource consumption, and environmental impact. Although existing pruning techniques offer certain advantages, they frequently entail trade-offs, such as reduced model accuracy, increased computational overhead, and inadequate consideration of sustainability. To address these challenges, in this study we propose a conceptual framework aimed at improving the efficiency and sustainability of BERT models through innovative pruning techniques. This framework incorporates a novel multi-objective optimization algorithm that clarifies the trade-offs among model accuracy, computational expenses, inference speed, and energy usage. Furthermore, the study proposes an Ensemble of Winning Tickets, grounded on the lottery ticket hypothesis, for the pruning of BERT attention heads. Pruning techniques reduce model size while preserving performance, enabling the deployment of highly efficient and environmentally sustainable models. The experimental evaluations show how these techniques are effective to real world tasks including linguistic acceptability and sentiment analysis using the Stanford Sentiment Treebank (SST-2) and Corpus of Linguistic Acceptability (CoLA) datasets. The results show that the proposed multi objective method achieves significant model compression of BERT Base model being pruned to 70.4 million parameters gaining an accuracy of 96.3% on SST-2 dataset and F1-score of 95.2%. Furthermore, the pruned model demonstrated F1 score of 82.3% and 56% of Matthews Correlation Coefficient in CoLA dataset respectively. The findings of this study also include model size that has been reduced by 36%, with our ensemble model reaching greater performance as compared to the baseline BERT model on both SST-2 and CoLA datasets. The results show F1-score of 94% and 96%, respectively, and accuracy scores of 95% and 96% on the two datasets. The findings of this study demonstrate that advanced pruning techniques enhance model performance, computational efficiency, and environmental sustainability. This study present practical framework to researchers and practitioners to move from large, resource-heavy models to sparse, efficient, and scalable options, thereby aiding the broader initiative to implement deep learning models in an environmentally conscious way.Item type:Item, An intelligent energy efficiency framework in cloud computing using deep reinforcement learning(Botswana International University of Science and Technology, 2025-01) Puso, Nomsa; Sigwele, Tshiamo; Mpoeleng, DimaneThe rapid expansion of cloud computing has resulted in a large increase of energy consumption, posing serious environmental and economic concerns. Data centers consume about 7% of worldwide electricity, a number that is anticipated to climb to 13% by 2030, contributing 2% of global emissions, comparable to the aviation industry. Current frameworks face challenges such as high energy consumption, poor job scheduling, non-adaptability to dynamic environments, and model optimization issues. These limitations demand the development of more efficient and adaptable solutions. This research proposes an intelligent energy efficiency framework for cloud computing based on Deep Reinforcement Learning (DRL). By dynamically adjusting to real- time workloads, this framework is designed for optimizing the energy consumption while maintaining the quality of service (QoS). This proposed DRL model formulates job scheduling through state and action spaces, incorporating real-time job characteristics such as arrival rates, types, and deadlines. The framework includes decision-making phases and reward functions to ensure effective and adaptable scheduling. The results show that the proposed DRL framework significantly outperformed related works demonstrating a 38% reduction in cloud energy consumption, a 98.4% jobs success rate, an improvement in QoS of 115%, a 66% reduction in response times, and a resource utilization of 90.7%. Additionally, the framework's scalability and adaptability to dynamic cloud conditions were validated through rigorous testing under various scenarios, confirming its robustness in handling fluctuating workloads and diverse job types. These outcomes address the identified problems by significantly enhancing energy efficiency, job scheduling effectiveness, and overall system performance.