Student Performance Prediction Using Machine Learning

This paper investigates the application of supervised machine learning algorithms for early student performance prediction and failure-risk classification. Driven by the proliferation of educational data within Learning Management Systems (LMS), early-warning systems are critical for identifying at-risk students before formal midterm or final evaluations occur. Utilizing a comprehensive dataset of 5,000 student records spanning demographic, academic history, behavioral engagement, and assessment scores, the target variable was collapsed into a binary outcome separating "At Risk" from "Not At Risk" students. Four classification models—Decision Tree, Gaussian Naive Bayes, Logistic Regression, and Random Forest—were trained, tuned, and evaluated on a stratified held-out test set. Experimental results indicate that Random Forest achieved the highest overall accuracy (92.30%) and F1-score (0.9272), whereas Logistic Regression yielded the highest ROC-AUC (0.9711) and superior recall on the minority at-risk class (0.915). Furthermore, feature importance analysis reveals that historical and ongoing academic performance indicators carry substantially more predictive weight than behavioral or demographic features. This study highlights the fundamental trade-off between aggregate accuracy and error-cost sensitivity when deploying early-warning systems in higher education.

17 Aug 2026 0 225