Enhancing Dropout Prediction with RUSBoost and BalanceCascade Algorithms: Tackling Class Imbalance in Real-World Educational Data in South Korea

Class imbalance presents a significant challenge in machine learning, especially in educational data analytics, where minority class instances are often critical. This study compares the performance of two advanced techniques, RUSBoost and BalanceCascade, for addressing class imbalance using real-world educational datasets in South Korea. We utilized data from the Korean Educational Longitudinal Study (KELS) from 2013 to 2021, focusing on 4,385 first-year university students in 2021. The datasets were preprocessed and categorized into various factors, including personal, family, and school factors. We implemented RUSBoost and BalanceCascade and evaluated their performance using four different base learners: Random Forest, Support Vector Machine (SVM), k-Nearest Neighbors (k-NN), and Logistic Regression. The models were assessed using multiple performance metrics, including Area Under the Receiver Operating Characteristic Curve (A-ROC), Precision-Recall Curve (A-PRC), Kolmogorov-Smirnov (K-S) statistic, and F-measure. RUSBoost demonstrated superior performance across most metrics and datasets, particularly excelling in A-ROC and K-S statistics. It consistently outperformed BalanceCascade, showing its robustness and efficiency. BalanceCascade, while competitive, showed slightly lower performance, especially with A-PRC and F-measure metrics. The comparative analysis revealed that RUSBoost’s more straightforward and faster approach made it a more practical choice for handling class imbalance in the educational sector. The findings suggest that RUSBoost is a highly effective method for improving classification performance on imbalanced datasets. Its simplicity and efficiency suit real-world applications, including educational data analytics. Future research should explore further enhancements to these techniques and their applicability in other domains. This study provides valuable insights into selecting appropriate methods for class imbalance, contributing to developing fair and accurate predictive models.