Research Publications
Permanent URI for this communityhttps://rda.sliit.lk/handle/123456789/4194
This main community comprises five sub-communities, each representing the academic contribution made by SLIIT-affiliated personnel.
Browse
111 results
Filters
Advanced Search
Filter by
Settings
Search Results
Publication Open Access Solar Hotspot Detection Using VHDL-Simulated Fixed-Point SVM: A Methodology Toward FPGA Realization(Faculty of Engineering, 2026-03) Fernando, N; Seneviratne, L; Weerasinghe, N; Rathnayake, N; Hoshino, YEarly detection of thermal hotspots in photovoltaic modules is critical to ensuring their efficiency, safety, and longevity. This study presents a complete end-to-end methodology for implementing a fixedpoint Medium Gaussian Support Vector Machine classifier using VHDL for a Field Programmable Logic Array. The approach begins with feature extraction from thermal images of healthy and defective solar panels, which focuses on MPEG-7 descriptors. The study shows that high impact for hotspot detection comes from blue chrominance contrast. A medium Gaussian SVM model is trained in MATLAB and converted to a fixed-point Q1.15 format for hardware compatibility. Key parameters, including support vectors, Lagrange multipliers, bias, and kernel scale, are extracted and verified in a custom Python environment to ensure numerical alignment with MATLAB results. The validated model is then implemented in synthesizable VHDL. It is verified using GHDL and the GNU Tool Kit waveform viewer, confirming bit-accurate hardware behaviour. Results show classification accuracy exceeding 99.3% with negligible performance loss due to quantization. The design achieves deterministic latency through an FSM-based structure and parallel feature processing for a 300-support vector and 222-feature system. This method enables low-power, real-time inference on a UAV-based edge platform, primarily focusing on drones.Item Embargo Interactive Sinhala Letter Learning Module for School Children (Grade 1 to 5)(Springer Science and Business Media Deutschland GmbH, 2026) Weerasooriya, K; Udana, I; Jayasinghe, L; Kasiwaththa, J; Rajapaksha, S; Kumari, SSinhala is the native language of most people in Sri Lanka. However, most of the children find it difficult to write Sinhala letters fast and accurately, this may undermine their confidence and affect grades. The primary issue is that the parents usually lack their time in order to assist their children in their studying at home. Few interesting tools also exist to teach children how to write in Sinhala in an interesting and effective manner. To address these issues we have developed the ”Interactive Application of the Sinhala Language to School children (Grade 1 to 5) which is a web based application, to allow children studying in primary schools to enhance their knowledge of the Sinhala language. This app provides children an entertaining and effective method of learning how to write Sinhala letters. The system combines instructions in animation, touch tracing finger tools, hand writing recognition and immediate feedback such that kids can learn Sinhala writing, and the non touch screen users can post their written letters on paper to be analyzed individually as to feedback analysis. The system uses handwriting recognition to provide real-time feedback on accuracy and speed. The system also monitors progress and generates comprehensive reports to help children and parents in identifying areas requiring improvement. The application uses a combination of engaging letter tracing and intensive deep learning which are not present in other learning tools. Additionally, the system will aid parents to mentor their children in education even when they are in charged schedules and also enable children improve their skills in Sinhala writing. We offer to make the learning of Sinhala to school students in Sri Lanka easier, more relevant and interesting.Publication Open Access Machine Learning-Based Early Warning Systems for Urban Floods: A Case Study in Nilwala Basin(Faculty of Engineering, 2026-01) Abayapala A.I.; Lindamulla L.M.L.K.BThis study pioneers the integration of Graph Neural Networks (GNNs) into flood forecasting systems, extending the predictive horizon from short-term forecasts to 7 days by effectively capturing spatial dependencies between rainfall stations. Focusing on the flood-prone regions of Matara and Galle districts within the Nilwala Basin, the research addresses the limitations of conventional forecasting methods by leveraging historical hydrological data, including daily rainfall records from six key stations and flow data from Pitabeddara. A hybrid machine learning framework combining Random Forest (RF) and K-Nearest Neighbors (KNN) models was developed to predict river discharge using rainfall data, overcoming challenges posed by limited water level data. The inclusion of GNNs introduces a novel approach to modeling complex spatial relationships, enabling improved accuracy in long-term flood prediction, particularly during extreme events. The proposed system demonstrates significant advancements in predictive reliability, offering a timely and accurate early warning tool to enhance disaster preparedness and risk management in the Nilwala Basin. This research underscores the transformative potential of datadriven methodologies in addressing the challenges of flood-prone regions.Publication Open Access A Data-Driven Approach to Predicting Ischemic Heart Disease Risk in Monaragala: Integrating Lifestyle and Symptom Factors with Machine Learning(Faculty of Engineering, 2025-09-09) Meddepola, M.A.R.L.; Wickramasinghe, B.M.G.S.T.S.K.Ischemic Heart Disease (IHD) remains a leading cause of mortality worldwide and presents a critical challenge in underserved rural areas such as Monaragala, Sri Lanka. Traditional IHD prediction methods predominantly depend on clinical diagnostics like ECGs and blood tests, which are often unavailable or inaccessible in such regions. This study aims to bridge this gap by developing a machine learning-based prediction model that utilizes only lifestyle and symptom-related data, eliminating the need for invasive clinical procedures. A dataset comprising lifestyle habits (e.g., diet, smoking, alcohol use, exercise) and symptom indicators (e.g., chest pain, fatigue, dizziness) was collected via surveys. Feature selection using Logistic Regression identified the top eight most relevant predictors. Five machine learning algorithms, Logistic Regression, K-Nearest Neighbors, Support Vector Machine, Decision Tree, and Random Forest, were trained and evaluated. Among them, the Random Forest model achieved the highest performance with an accuracy of 83.5%, precision of 0.86, recall of 0.78, and F1- score of 0.81, demonstrating strong predictive capability based solely on non-clinical features. In addition, a web-based self-assessment tool was developed to make the model accessible to the public, particularly targeting individuals in rural areas with limited healthcare access. The tool enables users to input basic lifestyle and symptom information and receive a real-time risk assessment. The findings confirm that the model leveraging lifestyle and symptom data can effectively identify individuals at risk of IHD. This approach supports the development of scalable, low-cost, and user-friendly screening tools that can enhance early detection and preventive care, especially in rural and resource-constrained settings.Publication Open Access Predictive Modeling for Personalized Cancer Therapy Using Reinforcement Learning(Faculty of Engineering, 2025-09-09) Edirisinghe M.M; Gunarathne,J H M S MAdaptive therapy is transforming cancer treatment by enabling dynamic, patient-specific interventions that adapt to tumor progression and individual variability. Unlike traditional fixed-dose regimens, adaptive therapy leverages the evolutionary dynamics of tumors to extend treatment effectiveness and delay resistance. Reinforcement Learning (RL), an area of artificial intelligence focused on sequential decision-making, offers a robust framework for optimizing these adaptive strategies. RL can learn optimal treatment policies by interacting with computational models of tumor growth and drug response, continuously adjusting regimens based on observed tumor states, resistant cell populations, and biomarkers. This approach allows for the creation of personalized therapies that maintain long-term tumor control while minimizing toxicity and the emergence of resistance. The integration of RL into predictive modeling for cancer therapy represents a paradigm shift, enabling smarter, safer, and more effective treatments that are dynamically tailored to each patient’s evolving disease. This paper reviews the foundational concepts of adaptive therapy and RL discusses tumor modeling approaches, examines RL algorithms, and addresses current challenges and future directions in the field.Publication Open Access Enhancing Healthcare Predictive Models Through Privacy- Preserving Synthetic Data Generation(Faculty of Engineering, 2025-09-09) Edirisinghe M.M; Gunarathne J.H.M.S.M; Wanniarachchi W.A.A.MThe advancement of healthcare predictive modeling is closely tied to the availability and quality of patient data. However, privacy regulations and ethical concerns often hinder data sharing, making it a persistent challenge. As a solution, privacy-preserving synthetic data generation has emerged, enabling the creation of artificial datasets that retain the statistical properties of real data while protecting individual privacy. This paper explores the use of such synthetic data throughout the clinical risk prediction pipeline by leveraging state-of-the-art generative models. We evaluate their utility in exploration data analysis, feature selection, model training, and deployment. Our study focuses on synthetic data generated using advanced models such as Differentially Private GANs (DPGAN), Private Aggregation of Teacher Ensembles GANs (PATEGAN), and Anonymization through Data Synthesis GANs (ADSGAN). Using these techniques, we created synthetic versions of the UK Biobank ever- smoker cohort. These synthetic datasets were shown to reproduce key statistical patterns, support effective feature selection, and enable accurate lung cancer risk prediction modeling all without using real patient data. We compare synthetic data with other privacy-enhancing technologies like federated learning and highlight a key advantage: synthetic data allows the direct use of existing analytical and machine learning tools without modification. Additionally, we examine deployment models such as "no- release" and "delayed-release," emphasizing how synthetic data can speed up research and enable broader data sharing while maintaining GDPR compliance. Overall, this study demonstrates the potential of synthetic data to transform healthcare research, software testing, education, and collaboration while carefully navigating the trade-off between privacy and utility.Item Open Access An integrated data-driven approach for Chronic Kidney Disease of Unknown Etiology (CKDu) risk profiling and prediction in Sri Lanka(SPIE, 2025) Rajapaksha, N; Rajawasan, H; Ubeysinghe, R; Perera,S; Swarnakantha, N.H.P.R.S; Gamage, M; Nanayakkara, N; Wijayakulasooriya, J; Herath, D; Lakmali, MChronic kidney disease of unknown etiology is a significant public health issue in Sri Lanka, especially in rural farming communities. The exact causes remain unclear, with potential links to environmental and socio-economic factors. This research employs Biological Data and Geographic Information Systems to analyze risk factors such as water quality, agricultural practices, climatic conditions, Demographic Factors, Socio-economic Factors. This study uses data from government health records, the Centre for Research-National Hospital Kandy, and field surveys. By identifying patterns and correlations, the study aims to inform public health interventions and reduce the impact of CKDu, ultimately improving health outcomes for affected populations. This will greatly contribute to preventing the disease, reducing the risk, and identifying patients at an early stage.Item Embargo Performance Analysis of Text Classification Algorithms for Dhivehi Language Documents(Institute of Electrical and Electronics Engineers Inc., 2025) Mohamed, F.R; Haddela, P.SThis study examines the effectiveness of various machine learning algorithms in classifying text written in 'Dhivehi,' the official language of the Maldives. As a low-resource language with limited research in text analytics, 'Dhivehi' poses unique challenges due to its distinctive linguistic properties. To address these challenges, this research evaluates the performance of algorithms, including Support Vector Machines, Naive Bayes, Decision Trees, Neural Networks, XGBoost, and Random Forest, leveraging a newly curated 'Dhivehi' language dataset. The evaluation highlights that K-Neighbors achieved the highest performance, with an accuracy of 64.7% and F1 scores (macro: 0.640, weighted: 0.642), demonstrating a strong balance between precision and recall. Support Vector Machines (accuracy: 63.9%) and XGBoost (accuracy: 62.8%) also showed competitive results, with SVM slightly outperforming XGBoost in F1 metrics. Decision Tree exhibited the lowest performance across all metrics. The findings provide critical insights into improving text classification for low-resource languages and contribute to developing natural language processing tools adapted explicitly for 'Dhivehi.' Furthermore, the dataset is publicly available on Mendeley data under the name 'Dhivehi Categories data set' to foster future research and innovation in this domain.Item Embargo AI Interviews with Facial Emotion Recognition for Real-Time Feedback and Career Recommendations(Institute of Electrical and Electronics Engineers Inc., 2025) Herath R.P.N.M; Arachchi D.S.U.; Gunaratne M.H.B.P.T.; Hansana K.T.; Wijayasekara, S.K; Jayasinghe, DThe hiring process is complex, requiring evaluation of candidates across multiple dimensions, including technical proficiency, behavioral traits, and credibility. Traditional interviews often suffer from biases and inefficiencies. This research presents an AI-driven Interview System integrating Machine Learning (ML), Natural Language Processing (NLP), and Computer Vision to automate and enhance recruitment. The system generates contextual interview questions, evaluates candidate responses using LLM-based scoring models, and provides real-time feedback for engagement. It includes speech-to-text transcription and offensive word detection to ensure professionalism. The behavioral analysis module leverages facial emotion recognition and computer vision to assess non-verbal cues such as confidence and attentiveness. Additionally, Curriculum Vitae (CV) parsing and LinkedIn data extraction use NLP-based entity recognition to extract educational background, work experience, and key skills, enabling personalized interviews. The technical assessment module administers real-time coding challenges, evaluating solutions for correctness, efficiency, and best practices while providing AI-generated feedback. By automating these key hiring aspects, this system enhances objectivity, efficiency, and decision-making, ensuring a data-driven, unbiased, and scalable selection process while improving the candidate's experience and employer insightsItem Embargo Blockchain-Based Custody Evidence Management System for Healthcare Forensics(Institute of Electrical and Electronics Engineers Inc., 2025) Jayasinghe R.D.D.L.K; Sasanka M.W.K.L; Athukorala D.A.S.M; Sandeepani M.A.D; Jayakody, A; Senarathna, AAs digital evidence increasingly growing in significance in healthcare forensics, safeguarding sensitive medical data's confidentiality, integrity, and limited access remains to be an important issue. Existing forensic evidence management systems are subject to data breaches and illegal access since they frequently lack significant privacy-preserving measures. In order to overcome such challenges, this research suggests a Blockchain-Based Custody Evidence Management System for Healthcare Forensics, which combines blockchain technology, machine learning, and encryption methods to improve security, privacy, and accessibility. To ensure accurate and efficient gathering of information, machine learning algorithms are used to extract handwritten and printed text from medical photographs. AES encryption ensures safe storage, while Fully Homomorphic Encryption (FHE) is used for dynamic access level control to protect gathered evidence. Identity verification is made possible via a web-based authentication system that uses Zero-Knowledge Proofs (ZKP) to protect privacy by preventing the disclosure of personal data. By preventing unintended modifications, blockchain technology is used to preserve the custody chain's integrity. Furthermore, machine learning-driven PII detection and masking methods balance the requirement for forensic investigation with privacy compliance by controlling data accessibility according to access entitlements. Based on permitted access levels, the system makes it possible to share safe evidence with law enforcement agencies, such as courts, the police, and other forensic groups. Using blockchain to guarantee data immutability, cryptographic security to restrict access, and artificial intelligence (AI) to safeguard data, this approach enhances the privacy, security, and dependability of handling forensic evidence in medical investigations