Enhancing Academic Performance Prediction: A Comprehensive Comparison of Machine Learning Techniques and Metrics

This study explores the prediction of academic performance by comparing various machine learning (ML) techniques across secondary Mathematics and Portuguese datasets. It examines ten ML models using binary and 5-grade classifications to identify at-risk students and assesses metrics such as Accuracy, Precision, Recall, F1-score, and Area Under the Curve. The results reveal that while Accuracy might align with other metrics in balanced datasets, it can be misleading in imbalanced ones. Locally Weighted Learning and Bagging excelled in Mathematics classifications, whereas Naïve Bayes and Logistic Regression were more effective in identifying at-risk students in Portuguese classifications despite lower Accuracy scores. The study underscores the importance of using multiple metrics to comprehensively evaluate model performance, particularly in imbalanced datasets. As a result, it highlights the potential of ML techniques to enhance educational interventions and support systems, thereby improving student outcomes and decision-making.