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
5 results
Search Results
Item Embargo Enhancing Chronic Kidney Disease Prediction : A Hybrid Approach Combining Logistic Regression and Random Forest Models(Institute of Electrical and Electronics Engineers Inc., 2025) Jathunga, T; Abeygunawardena, NThis study investigates the use of Machine Learning (ML) models for Chronic Kidney Disease (CKD) prediction, comparing Logistic Regression with L1 and L2 regularization, Random Forest , and a Hybrid Voting Classifier. The models were evaluated using performance metrics including accuracy, precision, recall, and F1-score, with the hybrid model demonstrating the highest accuracy of 99 percent, followed by Random Forest at 98 percent. Logistic Regression models achieved accuracies of 97 percent and 98 percent , with slight variations in recall for different classes. Cross-validation and learning curve analyses indicated minimal overfitting in ensemble models. These results emphasize the potential of ML models for accurate CKD prediction, suggesting further research into model optimization and data preprocessing techniques.Item Embargo Developing Predictive Models for Future Stress Likelihood and Recovery Time Using Behavioral and Emotional Data(Institute of Electrical and Electronics Engineers Inc., 2025) Weerasinghe W.P.D.J.N; Gunasekera H.D.P.M; Wickramasinghe B.G.W.M.C.R; Jayathunge K.A.D.T.R; Wijesiri, P; Dassanayake, TStress has a serious impact on mental and physical well-being, but treatments as usual are often unavailable and not effective over the long term. The AyurAura application combines imaginative Ayurvedic therapies with modern AI techniques to deliver customized stress reduction by way of Mandala art and music. This research develops two predictive models for the application. In its first model, the stress prediction probability is estimated from users' behavior in a questionnaire and the result can be used to proactively intervene. The second model forecasts time needed for recovery into a stress-free state by using the changes in daily emotional state and participation in app activities. Machine learning algorithms are used to prepare behavioral and emotional data for improved prediction performance. Trained on multi-institution datasets, both models delivered 90-95% accuracy, enabling the user to detect behavior eliciting stress and the degree needed for recovery. These results highlight the possibility of combining conventional therapeutics with contemporary tech for ongoing, affordable stress relief interventions with personalized needs in mind.Publication Open Access Forecasting Monthly Electricity Consumption for Energy Planning and Policy Development in Sri Lanka(Department of Mathematics and Statistics, Faculty of Humanities and Sciences, SLIIT, 2025-10-10) Pathirana, D; Arachchige, C. N. P. GFor effective energy planning, grid stability, and policy development especially in emerging nations like Sri Lanka accurate electricity consumption projections is essential. The goal of this project is to create a reliable model that can forecast the Ceylon Electricity Board's (CEB) monthly electricity consumption using a large dataset that includes macroeconomic variables, market indicators, peak demand, energy generation sources, and weather data. Autoregressive Distributed Lag (ARDL), Random Forest, and eXtreme Gradient Boosting (XGBoost) are the three models whose redicting performance is compared in this study. The most pertinent predictors were chosen using Recursive Feature Elimination with Cross-Validation (RFECV). Although XGBoost performed well throughout training, overfitting was a problem. ARDL was interpretable, however it was unable to detect longterm cointegration and could not represent non-linear connections. With the best accuracy and dependability on the test dataset without overfitting, Random Forest turned out to be the best model whereas Monthly Sales by Tariff in LKR, Fuel Cost by Power Stations in LKR, Electricity Generation from Thermal Coal in CEB (Gwh), Electricity Generation from Mannar Wind in CEB (Gwh), Day Peak Demand (MW), Night Peak Demand (MW), Average Monthly Rainfall (mm), and Gross Domestic Product (GDP) in LKR were the eight most important factors that were found to be involved in forecasting electricity consumption. On the test dataset, Random Forest, the best model chosen, had an accuracy of 77.34%, a Mean Absolute Percentage Error (MAPE) of 22.67%, a Root Mean Square Error (RMSE) of 12.62, and a Mean Absolute Error (MAE) of 11.11. However, the models might not be able to reflect long-term structural changes like the switch to electric vehicles or widespread adoption of renewable energy sources, and the study did not account for new elements like government policy reforms or energy efficiency initiatives. Nevertheless, the results show that machine learning, in particular Random Forest, can improve Sri Lankan electricity consumption predictions to aid in sustainable energy planning and policy choices.Publication Open Access Classifying Credit Card Payment Risk among Senior Citizens using Machine Learning on Limited Data(Department of Mathematics and Statistics, Faculty of Humanities and Sciences, SLIIT, 2025-10-10) Jayathilaka, G. A. M. K.; Ekanayake, E. M. P.; Appuhami, P. A. D. A. N.Sri Lanka is experiencing a significant demographic transition in view of lower fertility, enhanced life expectancy, and international migration, all of which have accounted for a higher proportion of senior citizens. The credit risk associated with a rising percentage of elderly population demands an investigation into the payment habits of senior citizens using credit facilities, for their own benefits and the sustainability of financial institutions. Addressing this issue, machine learning techniques are employed in this study in order to develop a viable model for classifying the credit card payment riskposed by the senior citizens based on their demographic and financial metrics at a leading private bank in Sri Lanka. Predictive dual-category classifications comprising of on-time payers and risky payers that include both late payers and dormant users are achieved using the machine learning algorithms of Logistic Regression, Random Forest, Support Vector Machine, Naïve Bayes, Extreme Gradient Boosting, and Multi-Layer Perceptron Neural Networks. Hyperparameter tuning and model performance ptimization were accomplished using Grid Search Cross-Validation, with models beingassessed by Accuracy, Precision, Recall, F1 Score, and AUC-ROC. Of those modelling techniques, the Random Forest excelled with 85% Accuracy, 85% Precision, 85.45% Recall, 84.99% F1 Score, and a high AUC-ROC score of 88 in classifying the risk posed by senior citizens, identifying payment settlement, average monthly credit card spending, average monthly debit, and number of credit card transactions as the key contributing variables. This classification method could be recommended for financial institutions with limited databases in setting credit limits tailored to senior citizens’ payment behaviour, reducing risk and promoting sustainable financial management for seniors.Publication Embargo Cricket Shot Image Classification Using Random Forest(IEEE, 2021) Devanandan, M; Rasaratnam, V; Anbalagan, M. K; Asokan, N; Panchendrarajan, R; Tharmaseelan, JCricket is one of the top 10 most played sport across the world regardless of age and gender. However, learning cricket has been quite challenging as the majority of the cricket-playing individuals are unable to afford quality infrastructure. While this has opened up many research opportunities to provide solutions to automatically learn cricket, very little work has been done in this era. In this paper, we focus on the batting skills of cricket players. We develop a Random Forest model to classify the cricket shot images using human body keypoints extracted with MediaPipe. Experiment results show the proposed model achieves an F1-score of 87% and outperforms the existing solution in a 5% margin. Further, we propose a similarity estimation approach to compare the user’s cricket image with popular international cricket players’ cricket shot images of the same type and retrieve the most similar one. The mobile application we developed based on our solution will enable cricket-playing individuals to analyze, improve and track their batting performances without the need of having a coach.