Student Performance Prediction Using Machine Learning Techniques for Early Academic Intervention

Identifying weak students only after the final examination leaves little room for corrective action. Machine learning (ML) can analyse attendance, internal marks, assignment records, previous academic results and study behaviour to flag at-risk students while the course is still running. This paper reviews ten recent studies (2025–2026) on student performance, dropoutrisk and early-warning prediction, covering classical classifiers, boosting ensembles, hybrid and stacked models, fuzzy decision support, deep learning and graph-based approaches. The studies are compared with respect to datasets, preprocessing, algorithms, evaluation metrics and reported outcomes. Recurring limitations are identified: small single-institution datasets, class imbalance, target leakage, optimistic validation, limited interpretability and a weak link between prediction and actual intervention. Based on these findings, a decision-support framework is proposed in which Logistic Regression, Decision Tree and Random Forest models are trained on academic and behavioural data, students are placed into risk tiers, and teachers receive alerts through a dashboard. The framework is designed to support, not replace, teacher judgement. Directions for future work are also outlined.