Mitigating Imbalanced Classification Problems in Academic Performance with Resampling Methods

Abstract

—The imbalanced dataset is a common problem in the educational performance environment, where the number of students with poor performance is much less than those who perform well. This can create problems when predicting academic performance using machine learning algorithms, which assume that the datasets have a balanced distribution across all classes. We compared three resampling methods: SMOTE, Borderline SMOTE, and ADASYN, and used five different classifiers (Logistic Regression, Support Vector Machine, Naïve Bayes, K-Nearest Neighbor, and Decision Tree) on three imbalanced educational datasets. We used five-fold cross-validation to assess two performance indicators: accuracy and recall. Although accuracy indicates the overall performance, we focus more on recall values because it is more incumbent to identify poor-performing students so that necessary interventions can be executed promptly. Our results showed that when resampling improved recall values, ADASYN outperforms SMOTE and Borderline SMOTE consistently, better classifying the poor-performing students. Overall, our results suggest that resampling methods can be effective in addressing the problem of imbalanced classification in academic performance. However, the choice of resampling method should be carefully considered, as the performance of different methods can vary depending on the classifier used.