Early Prediction of Student Academic Performance: A Critical Narrative Review of Machine Learning Methods, Evaluation Design, and Responsible Deployment Copy title

Abstract

This critical narrative review examines machine-learning approaches to the early prediction of student academic performance, focusing on logistic regression, decision trees, and random forests. It discusses the importance of dataset selection, preprocessing, model evaluation, data leakage, interpretability, algorithmic fairness, and reproducibility in educational data mining. The review also outlines a proposed framework for comparing predictive models using publicly available student-performance datasets. Particular attention is given to the limitations of small datasets and the ethical implications of using predictive analytics in educational settings. This paper synthesizes existing research and provides methodological recommendations for future studies. No new experimental results are presented.