Comparison of Classification Algorithms for Digital Business Students’ Academic Performance: SVM, Random Forest, XGBoost, and LightGBM with Class Imbalance Handling Using SMOTE

Abstract

This study aims to compare the performance of classification algorithms, namely Support Vector Machine (SVM), Random Forest, XGBoost, and LightGBM, in predicting the academic performance of Digital Business students at ISB Atma Luhur by handling class imbalance using the Synthetic Minority Oversampling Technique (SMOTE). The dataset consisted of 326 student records with 55 questionnaire-based Likert-scale features, GPA, and semester data classified into two academic performance classes. The research stages included data preprocessing, normalization, SMOTE implementation, feature selection using feature importance, model training, and evaluation using accuracy, precision, recall, F1-score, F1 Macro, AUC-ROC, and training time metrics. The results showed that the XGBoost algorithm achieved the best performance with an accuracy of 0.8621, an F1 Macro score of 0.85, and an AUC value of 0.91. LightGBM produced performance close to XGBoost while providing faster training time. The implementation of SMOTE successfully improved minority class classification performance across all algorithms, particularly in terms of F1-score. The findings indicate that the combination of boosting algorithms and class imbalance handling techniques is effective for machine learning-based academic performance prediction systems.