MSc in Information Technology
Permanent URI for this collectionhttps://rda.sliit.lk/handle/123456789/2484
Students enrolled in the MSc in Information Technology programme are required to submit a thesis as a compulsory component of their degree requirements. This collection features merit-based theses submitted by postgraduate students specialising in Information Technology. Abstracts are available for public viewing, while the full texts can be accessed on-site within the library.
Theses and Dissertations of the Sri Lanka Institute of Information Technology (SLIIT) are licensed under a
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License
.
Browse
102 results
Filters
Advanced Search
Filter by
Settings
Search Results
Publication Open Access Automated Analysis of Commenting Styles and Documentation Practices: A Data-Driven Approach to Software Quality and Maintainability(Sri Lanka Institute of Information Technology, 2025-12) Sathyangani, K.A.H.P.Software maintainability is strongly influenced by the quality of code comments, which guide developers in understanding system functionality and behaviour. Poorly written, missing, or ambiguous comments reduce productivity and increase the cost of maintenance. The current study introduces an automated, data-driven approach to evaluating comment quality in Java projects. The proposed solution, implemented as a Java-based tool named Comment Quality Analyser, automatically scans source files, extracts comments, and evaluates them using four quality dimensions: grammatical correctness, readability, understandability, and meaningfulness. The tool integrates LanguageTool for grammatical analysis, the Flesch Reading Ease metric for readability, heuristic rules for understandability, and a Jaccard-similarity-based algorithm for measuring semantic alignment between comments and code identifiers. The results are presented through JSON reports and an interactive HTML dashboard that visualises the quality distribution across files. Real-world validation was conducted using the Apache Commons IO open-source repository, containing over 100 comments. Experimental results indicate that the system provides consistent scoring with an average accuracy of 86 % when compared with manual reviews. The proposed framework contributes to improving software documentation practices and offers a foundation for further research integrating Natural Language Processing (NLP) and Machine Learning (ML) to enhance software maintainability analysis.Publication Open Access Sri Lankan Sign Language Detection System for Service Robots(Sri Lanka Institute of Information Technology, 2025-12) Rajamanthri, H. USinhala Sign Language (SSL) is the main way many Deaf and hard-of-hearing people in Sri Lanka communicate, but there are very few digital tools to recognize it automatically. This thesis addresses that problem by developing deep learning models that can recognize SSL gestures accurately while also working in real time. A dataset of 5,000 SSL videos covering 50 gestures was created and divided into training, validation, and test sets, with preprocessing steps such as frame resizing, normalization, and data augmentation. Four models were tested: a CNN to learn visual features, an LSTM to capture movement over time, a combined CNN-LSTM model, and YOLOv8 for fast, detection-based recognition. The CNN achieved the best overall performance with 98.32% accuracy, followed closely by the LSTM at 98.18%, while YOLOv8 reached 97.20% accuracy but delivered very high real-time speed of up to 90 frames per second. The CNN-LSTM model performed less well than expected. A working prototype was also built, allowing users to perform signs and see the recognized output in Sinhala or English, with optional speech, showing the system’s practical value for inclusive communication.This research establishes a foundational framework for SSL recognition, showing that deep learning models can achieve state of the art accuracy while highlighting the trade offs between accuracy, latency, and deployment feasibility in real world contexts.Publication Embargo Representing Music Visually for the Hearing-Impaired using AI(Sri Lanka Institute of Information Technology, 2026-01) Varaprasath,S.This research presents the design and implementation of an AI-based music visualization system that enables Deaf and Hard-of-Hearing (DHH) individuals to experience music through dynamic visual and haptic feedback. The system integrates both classical machine learning and deep learning models to analyze key audio features, such as tempo, pitch, and spectral energy, and lyrical sentiment, translating them into synchronized visual elements like color, motion, and shape. The visualization framework employs a combination of CNN–LSTM architectures for emotion detection and SVM-based genre classification, ensuring accurate mapping between musical attributes and emotional states. A Flask-based web interface was developed to deliver real-time, adaptive visualizations generated from user-uploaded songs. The study further proposes structured design guidelines and best practices for inclusive audiovisual systems, derived from cross-disciplinary literature in affective computing, accessibility design, and cognitive psychology. Experimental evaluation demonstrates that the system effectively translates musical emotion into coherent and perceptually comfortable visual patterns, achieving high usability and accessibility for DHH users. Beyond accessibility, the research contributes a conceptual framework for emotion-aware, AI-driven multimedia systems, offering applications in interactive art, music education, and assistive technology. The findings highlight the potential of artificial intelligence to bridge sensory gaps and make musical experiences universally inclusive.Publication Open Access Real-Time ML Integration with CFS to Improve Power Efficiency in Non-Hybrid Linux Systems(Sri Lanka Institute of Information Technology, 2025-12) Dissanayake M.D.Modern non-hybrid Linux systems rely on the Completely Fair Scheduler (CFS) and Dynamic Voltage and Frequency Scaling (DVFS) to balance performance and energy efficiency. However, this default approach has key limitations: CFS distributes tasks evenly across cores without distinguishing between workload types, while DVFS adjusts frequency based on aggregate load. As a result, CPU-bound and I/O-bound tasks often share the same cores, leading to unnecessary frequency boosts and wasted energy for tasks that don’t require high frequencies. This thesis proposes a machine learning-enhanced scheduling framework that integrates task-type awareness into non-hybrid Linux systems. A decision tree classifier predicts whether tasks are CPU-bound or I/O-bound using lightweight runtime features such as execution time, waiting time, context switches, priority, and niceness values. Classified CPU-bound tasks are consolidated onto selected cores running at higher frequencies, while I/O-bound tasks are grouped on separate cores maintained at lower frequencies. To prevent thermal hotspots and performance degradation, periodic rotation of CPU-bound and I/O-bound core groups is implemented. The scheduler is evaluated against the default CFS using controlled CPU-intensive, I/O-intensive, and mixed workloads generated with Sysbench and FIO. Energy consumption, performance, and Energy-Delay Product (EDP) are measured using perf and turbostat. Experimental results show that the ML-enhanced scheduler reduces energy consumption and improves EDP significantly, whilemaintaining performance levels comparable to the default scheduler. These findings highlight the potential of ML-based task classification to improve DVFS utilization and deliver more energy-efficient scheduling in general-purpose, non-hybrid Linux systems.Publication Open Access IoT - Based Multispectral Imaging System for Early Detection of Diseases, Pest Infestations in Kochchi (Dark Green Scotch Bonnet) Plants Using Deep Learning Algorithms(Sri Lanka Institute of Information Technology, 2025-12) Banneheka, B. M. C. P.Kochchi, also known as the Dark Green Scotch Bonnet, is a valuable chili variety that thrives in Sri Lanka. However, the yield and quality of this crop often suffer due to pests and plant diseases. Traditional detection methods rely on manual inspection, which is labor-intensive, subjective, and typically identifies symptoms only after visible damage occurs. This study proposes a multispectral imaging (MSI) framework combined with deep learning algorithms to achieve early and accurate disease detection under greenhouse conditions. A DJI Mavic 3 Multispectral drone was used to capture synchronized RGB and multispectral bands (Green, Red, Red-Edge, NIR). Preprocessing involved band alignment, vegetation index computation (NDVI, NDRE, GNDVI), and reduce noise, all aimed at spotting early signs of distress before they become severe. A convolutional neural network (CNN) was trained for binary classification of healthy versus infected plants, achieving high accuracy across cross-validation. For plants classified as infected, severity mapping was performed using advanced instance segmentation models, YOLOv11 and Mask R-CNN, followed by HSV color transformation and K-means clustering to quantify lesion area as a percentage of total leaf surface. Evaluation using Accuracy, Precision, Recall, F1-score, Intersection over Union (IoU), and Dice Coefficient confirmed the robustness of the system. The proposed approach demonstrates the potential of multispectral imaging combined with deep learning for early disease detection and severity quantification in Kochchi cultivation, reducing reliance on pesticides and supporting sustainable agriculture.Publication Open Access Instruction-Guided AI-Driven Predictive Analytics for User-Level CPU Optimization and Overutilization Detection in Cloud Infrastructure Management(Sri Lanka Institute of Information Technology, 2025-12) Rathnayaka, R.M.C.P.Cloud computing provides scalable and cost-efficient infrastructure, yet effective CPU resource management continues to pose challenges. Overutilization of CPU resources can result in degraded performance, violations of service-level agreements (SLAs), and increased operational costs. While many existing solutions focus on optimizing system- or virtual machine-level performance, there has been limited exploration of user-level workload variability, where CPU demand often fluctuates significantly across different tenants. To address this gap, the study introduces an AI-driven predictive analytics framework designed to manage CPU overutilization at the user level in cloud environments. The framework combines machine learning with an instruction-guided decision engine that generates actionable optimization strategies for administrators, rather than relying on full automation. This human-in-the-loop approach enhances transparency and trust in operational decisions. The evaluation used two datasets, including synthetic workloads that represented light, medium, and heavy users, as well as real-world traces from the Bitbrains GWA-T-12 dataset. Random Forest and Logistic Regression models were trained using features such as average CPU utilization and provisioned capacity. CPU overutilization was defined as usage exceeding 90 percent of the allocated CPU. The results indicated that the Random Forest model achieved higher predictive accuracy, surpassing 90 percent, and reached an AUC value of 0.99, outperforming Logistic Regression. The decision engine then translated these predictions into optimization instructions, including actions like workload migration and CPU scaling. Overall, the findings demonstrate that combining user-level prediction with instruction-guided optimization improves CPU resource management, reduces the risk of overutilization, and enhances cloud system performance. This research contributes a practical, lightweight solution that advances predictive cloud resource management while preserving administrative oversight and system interpretability.Publication Open Access Implementation of a Web-Based Complex Formula Editor(Sri Lanka Institute of Information Technology, 2025-12) RUBERU, T.D.B.S.Mathematical formula editing in web-based form has also not been easily achieved because it does not have real-time character-level editing, real-time handwriting recognition, and a smooth integration of the database. Current solutions like LaTeX editors or graphical tools that are not dynamically updated have the disadvantage of requiring one to re-write whole expressions to fix small mistakes. Moreover, they can only be accessible to students, researchers, and engineers to some extent due to the use of coding syntax or tedious interfaces. The proposed research is a new Web-Based Complex Formula Editor that would address these limitations. The system permits character level editing in real-time, so that users can edit individual formula portions directly. It is based on artificial intelligence (AI)-driven handwriting recognition to precisely change handwritten input into structured digital representations such as MathML and LaTeX. Besides, the editor has an equation retrieval system which is linked to the OpenMath standard. Through advanced web technologies and AI-based techniques, this performance fills the gap between usability and computing capabilities. The proposed editor has the capability of an easy to use, multipurpose platform to create and manage digital mathematical content and improve productivity and accessibility to a large extent.Publication Open Access Enhancing the Software Development Life Cycle through Integration of Generative Artificial Intelligence(Sri Lanka Institute of Information Technology, 2025-12) Thilakarathna W L DThe Software Development Life Cycle (SDLC) is a foundational framework in software engineering, yet its documentation process spanning Business Requirement Documents (BRDs), User Stories, Test Cases, and Automation Scripts remain highly manual, inconsistent, and resource intensive. This study presents a Generative Artificial Intelligence (GenAI)–driven framework designed to automate and integrate SDLC documentation from end to end, enhancing traceability, efficiency, and quality across all stages. The proposed system employs Retrieval-Augmented Generation (RAG) in combination with FAISS-based semantic retrieval and LangChain orchestration to extract structured requirements from BRDs, generate standardized User Stories, derive Test Cases, and produce executable Cucumber and Selenium scripts. Both qualitative and quantitative methodologies were adopted: interviews with Business Analysts (BAs) and Quality Assurance (QA) engineers identified documentation challenges, while experimental evaluation measured performance and accuracy. The results demonstrate that the framework reduces documentation time by over 90%, ensures 100% traceability between SDLC artifacts, and achieves over 95% accuracy in generated outputs. Compared to general-purpose LLMs such as ChatGPT and Gemini, the proposed approach delivers structured, consistent, and production-ready documentation with minimal human intervention. This research contributes a validated and scalable model for AI-assisted SDLC automation, offering a significant step toward intelligent, traceable, and self-sustaining software documentation pipelines. The findings have both theoretical and practical implications, supporting the broader integration of Generative AI within enterprise software engineering and quality assurance practices.Publication Open Access Enhancing Student Performance Prediction Using Real-Time Data and Explainable Artificial Intelligence in Higher Education(Sri Lanka Institute of Information Technology, 2025-12) Dewduni, P.H.H.C.Accurate prediction of student performance has become a strategic priority in higher education, driven by increasing learner diversity, rising dropout rates, and limited institutional resources. Traditional statistical models based on historical records are limited in capturing the dynamic and multidimensional nature of student learning. This study proposes a real-time predictive framework that integrates machine learning with explainable artificial intelligence to achieve both predictive accuracy and interpretability. Data were collected from 2175 undergraduate students from Faculty of Humanities & Social Sciences, University of Ruhuna across five course modules, combining institutional academic and learning management system records with a structured survey capturing financial and psychosocial indicators. After a systematic preprocessing pipeline, five machine learning classifiers were developed and evaluated, including Logistic Regression, Decision Tree, K Nearest Neighbours, Support Vector Machine, and Random Forest. Model performance was assessed using accuracy, precision, recall, F1 score, and area under the curve. Random Forest achieved the highest accuracy of 96.92%, while Support Vector Machine achieved the strongest discriminative capability with the highest area under the curve. To address interpretability, SHapley Additive exPlanations provided global feature attributions, and Local Interpretable Model Agnostic Explanations generated case-specific insights. Results consistently identified attendance and assignment performance as dominant predictors, while behavioural engagement and financial and psychological dimensions offered complementary contributions. A decision support dashboard was designed to operationalize these insights by enabling early identification of at-risk learners, explaining contributing factors, and generating individualized reports for timely intervention. The proposed framework demonstrates that robust predictive accuracy and interpretability can be achieved simultaneously, providing a practical tool for enhancing student retention, equitable support, and efficient resource allocation in higher education.Publication Open Access Edge-Aware Optimization for AI-Driven Anomaly Detection in Smart City Wireless Networks(Sri Lanka Institute of Information Technology, 2025-12) Kumarasiri, D.G.S.M.The rapid growth of smart cities has led to increased reliance on wireless networks and IoT devices, raising critical security concerns, particularly regarding anomaly detection. Traditional methods fall short in dynamic and resource constrained environments due to their limited adaptability and high computational demands. This paper presents a comprehensive review of AI-driven approaches to anomaly detection in smart city wireless networks, focusing on real-time adaptability and edge-aware optimization. It evaluates state of the art techniques such as federated learning, lightweight deep learning models, and hybrid AI frameworks that balance detection accuracy with efficiency. Public datasets like CICIDS 2017 and IoT-23 are discussed in detail for their role in training and evaluating these systems. Despite significant progress, existing models struggle with scalability, high false positive rates, and limited deployment readiness for edge devices. The paper proposes a conceptual framework to address these gaps and lays a foundation for future research in secure, scalable smart city networks. Additionally, this study incorporates architectural visualization and proposes feasible real-time deployment through edge-based simulators, addressing gaps in prior implementations and privacy safeguards.