ARCHIVES

Predicting student academic performance from prior academic records has become one of the most actively studied problems in educational data mining (EDM) and learning analytics, motivated by the need to identify at-risk learners early and to guide timely pedagogical intervention. This paper presents a systematic research study on the use of machine learning techniques to predict student academic performance using historical performance indicators such as previous grades, test scores, attendance, and study-related behavioural attributes. We review the theoretical foundations of educational data mining, survey major supervised learning algorithms applied to performance prediction — including linear and logistic regression, decision trees, random forests, support vector machines, k- nearest neighbours, naive Bayes, and artificial neural networks — and synthesise findings from twenty-five peer-reviewed and conference-published studies. We then propose and empirically evaluate a predictive framework on a literature-informed, simulated secondary-school performance dataset, comparing eight algorithms using accuracy, precision, recall, F1-score, and root-mean-square error as evaluation metrics. Results confirm that prior grades remain the single strongest predictor of final academic outcomes, that ensemble methods such as Random Forest and Gradient Boosting consistently outperform single classifiers, and that the inclusion of behavioural and socio-demographic features yields only marginal improvement once prior academic performance is accounted for. The paper concludes with a discussion of practical implications for early-warning systems, the ethical considerations of predictive analytics in education, and directions for future research, including explainable artificial intelligence and the integration of multimodal learning data.