Addressing Class Imbalance in Machine Learning for Predicting On-Time Student Graduation at The Islamic University of Riau

Timely graduation is an important indicator of academic performance in higher education. However, many students still fail to graduate on time, prompting the need for predictive models to support academic decision-making. This study aims to analyze the impact of class imbalance on machine learning algorithm performance in predicting student graduation at the Islamic University of Riau. Data were obtained through questionnaires and labeled into “graduated on time” and “not on time” classes, which were initially imbalanced. The Synthetic Minority Over-Sampling Technique (SMOTE) was applied during preprocessing to balance the dataset. Four machine learning algorithms were compared: Decision Tree, Gaussian Naive Bayes, K-Nearest Neighbors, and Support Vector Machine. The evaluation was conducted with and without SMOTE, using accuracy, precision, recall, F1-score, and confusion matrix. Results showed significant performance improvements after applying SMOTE, with all models achieving around 99% accuracy. SVM achieved the most stable results across both conditions. The study highlights the effectiveness of SMOTE in improving classification fairness and reliability, especially in datasets with class imbalance. This work may assist universities in early intervention for students at risk of late graduation.