Handling Imbalanced Dataset in Online Learning Performance Prediction Using Resampling Techniques

Data has emerged as one of the world’s most valuable resources, offering untapped potential through data mining to generate actionable insights. Similar to refining precious materials like gold or diamonds, data mining transforms raw information into meaningful knowledge. However, a significant challenge in this process lies in addressing imbalanced datasets, where the disparity between majority and minority classes often results in biased predictions and reduced model accuracy. This study investigates the effectiveness of various resampling techniques in handling class imbalance and enhancing predictive performance. Specifically, we evaluate Random Oversampling, ADASYN, SMOTE, Random Undersampling, NearMiss, and hybrid resampling approaches on a highly imbalanced online learning performance prediction dataset. Our findings reveal that certain resampling methods improve model precision, recall, and F1-score, offering actionable insights into mitigating data imbalance challenges. This research provides a foundation for advancing prediction models in educational data mining, with implications for improving online learning performance and related key performance indicators.