Proactive forecasting of student academic outcomes via Random Forest and cumulative learning data

Abstract

This study develops a Random Forest-based predictive framework to forecast course outcomes for information technology students at Binh Duong University. Using institutional academic records, key features such as course codes, instructor identifiers, and encoded student IDs were combined with demographic data through a standardised preprocessing pipeline. To ensure robust validation, the experimental design utilised a strict data split (64% for training, 16% for validation, and 20% for testing) alongside 5-fold crossvalidation for hyperparameter tuning. The final model achieved an exceptional overall accuracy of 0.974 and a Receiver Operating Characteristic (ROC) Area Under the Curve (AUC) of 0.982. The framework is designed to address both course outcomes and Course Learning Outcome (CLO) attainment. Crucially, it attained a recall of 0.838 for the minority “Fail” class, demonstrating strong practical value for early risk detection. Feature-importance analysis confirmed the dominant influence of course- and instructor-related variables alongside cumulative student histories, ensuring high interpretability. By providing reliable predictions of course outcomes, the framework supports academic advisers in designing timely interventions. Ultimately, this research highlights the significant potential of institutional data for advancing educational decisionmaking and academic management.