Predicting Student Academic Success Using Machine Learning: A Comparative Analysis of Behavioral, Digital Engagement, and Socio-Academic Factors

Abstract

Background of study: Universities increasingly combine administrative records and learning-management-system data to identify students who may need academic support, yet prediction models often rely on a single data source and provide limited insight into the incremental value of behavioral, digital-engagement, and socio-academic factors. Aims and scope of paper: This study compares feature groups and machine-learning algorithms for predicting university student academic success while emphasizing interpretable and responsible learning analytics. Methods: A retrospective multi-institutional classification design is specified. Five algorithms—logistic regression, decision tree, random forest, support vector machine, and XGBoost—are evaluated using stratified cross-validation. For manuscript-development purposes, results are demonstrated with a clearly labeled synthetic dataset (N = 1,248) and must be replaced with verified empirical data before submission. Result: In the illustrative analysis, the combined feature model outperforms single-domain models; XGBoost achieves the highest ROC-AUC (.891) and F1 score (.842). Prior GPA, assignment completion, attendance, LMS active days, and missed deadlines are the most influential predictors. Conclusion: Academic-success prediction improves when static socio-academic information is integrated with dynamic behavioral and digital-engagement indicators. Prediction should support timely human-led intervention rather than automated high-stakes decisions.