http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/4859 2026-05-19T17:24:24Z 2026-05-19T17:24:24Z Dilakshigan, A. Erandi, J.D.T Chandana, A.W.S. http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/5296 2026-05-19T09:44:46Z 2026-01-28T00:00:00Z

Predictive modeling for injury risk and performance analysis in collegiate and female basketball players Dilakshigan, A.; Erandi, J.D.T; Chandana, A.W.S. This paper examines a machine-learning model of injury prediction and performance assessment on female and collegiate-level basketball players through interpretable workload-based models. The dataset was gathered among the players in the university and district teams based on training logs, sprint tests (20 m, 30 m, 50 m), records of each weekly session, player survey, and team performance databases, including variables of ACWR (Acute-toChronic Workload Ratio is a metric that compares short-term training load to longer-term load), fatigue rating, minutes played, points per game, history of previous injuries, and environmental conditions. The data consisted of 180 player-weeks of data in two training cycles. The exploratory analysis revealed obvious changes in workloads and increased ACWR spikes in female athletes. Logistic regression was used to predict injuries and it was found that fatigue scores and ACWR were significant predictors with 78 percent accuracy and 0.74 recall. Performance prediction using linear regression revealed that training intensity, sprint performance and minutes played were a combination of variables that could explain 62 percent of the variance in player output. Transparency of the research was achieved by the interpretation methods that could be used in practical coaching. The results have shown that ACWR with the use of playerspecific factors can offer valuable information on the concept of workload management, injury probability downplay and inform evidencebased decisions on behalf of the underrepresented athletic groups. This study can be used in the creation of non-discriminatory and understandable sports analytics to match the digital performance tools of the next generation.

2026-01-28T00:00:00Z Theeksha, J.R.N. Kudagamage, U.P. http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/5295 2026-05-19T09:03:45Z 2026-01-28T00:00:00Z

Data-driven prediction of university admission cutoff marks in Sri Lanka using machine learning Theeksha, J.R.N.; Kudagamage, U.P. Admissions into universities in Sri Lanka are determined through the cutoff marks published annually in the University Grants Commission (UGC), although the procedure itself is not only unclear but opaque to students as well as counselors. Since the procedure is unclear, students end up making academic choices based on misleading information. This research proposes a machine learning model for accurate determination of the cutoff marks in the UGC university admissions. The proposed approach uses the UGC cutoff dataset (2020-2025), with features like the year, the district quota, the stream, the university, the degree, the intake capacity, and the cutoff marks. The final supervised regression approach using XGBoost with year-related features and lag features for cutoff points is proposed. In this work, the error in the regression method is calculated by the RMSE, MAE, and R² values. The regression analysis will be done by the subgroup error analysis for districts and streams, and the regression results will be explained by SHAP values. The RMSE, MAE, and R² score of the XGBoost algorithm come out to be 0.2062, 0.1446, and 0.7484, respectively. Among various factors given importance by the algorithm, Z score is given maximum importance, followed by subject stream, district quota, university, intake capacity, and previous year cutoff trends. On checking the equity of the algorithm, it is found that the disparity of errors is very low, which makes it a fair algorithm. This study has shown the viability and accuracy of using machine learning algorithms to predict the admission cutoffs of UGC admissions. This work fills the very significant research gap created by the absence of publications focusing on the educational data-mining domain. This work can form the basis of an effective and transparent decision support system that can help students, counselors, and policymakers make educated decisions regarding educational planning and access to higher education.

2026-01-28T00:00:00Z Rathnaweera, R.C.L.U. Somaweera, W.T.S. Sandaruwan, R.M.T. http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/5294 2026-05-19T08:27:36Z 2026-01-28T00:00:00Z

AI-powered predictive modeling and comparative machine learning analysis for improving hospital operational efficiency Rathnaweera, R.C.L.U.; Somaweera, W.T.S.; Sandaruwan, R.M.T. The study addresses the critical problem that Sri Lankan state-run hospitals have no data-driven predictive tools of patient movement and resource allocation that have led to longer patients waiting and poor bed utilization. This study was done in the face of problems such as long patient waiting time and ineffective bed management owing to manual operations. The study followed an organized machine learning pipeline, where 500 records of patients from 01/2023 to 12/2023 were used to train and test predictive models that would predict Length of Stay, Readmission and Resource Requirement. The most important algorithms were the Random Forest, Gradient Boosting, and XGBoost, and they were tested according to the cross-validation and hyper parameter optimization. Findings confirmed that XGBoost was superior to the other models in that it was able to manage the complex interactions between features effectively and the test accuracy of 82.7% with F1-score of 0.809 indicating readmission prediction. Whereas, the Mean Absolute Error of the model in predicting length of stay (LOS) was approximately 9 days against a mean LOS of 15 days. It was also found that clinical and demographic factors such as Infection condition type, age group to 41-60, and department Intensive Care Unit are the most powerful predictors due to feature analysis, which indicated that the clinical presentation and patient characteristics were stronger indicators of the decision-making process of bed management than administrative characteristics. This study has shown that decision-support systems can be utilized based on solid ground using AI predictive models to optimize work processes in resource limited healthcare environments.

2026-01-28T00:00:00Z Jayathilaka, K.M.D.P.S.D. Rubasinghe, T.D. Kumara, B.T.G.S. http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/5293 2026-05-19T05:46:52Z 2026-01-28T00:00:00Z

A machine learning-based evaluation of English-to-Sinhala translation: comparing Google Translate, large language models, and human translators Jayathilaka, K.M.D.P.S.D.; Rubasinghe, T.D.; Kumara, B.T.G.S. Reliable translation from English to Sinhala is still a great challenge for many sophisticated translation systems using Sinhala as a low-resource language. Although Google Translate is widely used for translationpurposes, recent breakthroughs in large language models such as ChatGPTand DeepSeek provide entirely new opportunities for translation tasks. This study proposes one of the first thorough comparative analyses of English-Sinhala translation systems compared with human translation, both qualitatively and quantitatively. Google Translate, ChatGPT, DeepSeek, and human translations done by native Sinhala speakers were compared for translation quality on a carefully prepared dataset of 150English sentences for general, technical, and academic purposes. Translation quality was compared using BLEU, METEOR, and COMETscores, in addition to human assessment of fluency, grammatical accuracy, and semantic translation quality done by qualified human raters using a prepared rubric with inter-rater reliability tests. Machine learning models were also prepared for predicting translation quality using language-basedpredictors for translation efficiency and translation system identification. The experimental results show that human translations were rated highest on all translation quality measures. Among the automatic translationsystems, LLM-based translation systems performed better on contextual understanding of complex sentences than Google Translate, whichperformed reasonably on simple inputs. Correlation tests showthat COMET correlates better with human translation quality than BLEUandMETEOR. Moreover, the prepared machine learning models were able todetect translation quality trends accurately for translation systempredictions, making these models promising for translation qualityassessment in lowresource language environments.

2026-01-28T00:00:00Z