Handling Imbalanced Fraudulent Transaction Data Using SMOTE-Tomek and Random Forest: A Classification Approach
Muhammad Ghifari Ilham, Adi Winarno, Moch. Lutfi, Artanti Indrasetianingsih · BEST Journal of Applied Electrical Science & Technology · 2025
This research aims to address the class imbalance problem in fraud detection using hybrid resampling techniques, specifically SMOTE-Tomek, combined with Random Forest classifiers. Imbalanced data in fraud detection tasks can severely hinder model performance, resulting in poor detection of minority (fraud) cases. By employing SMOTE to oversample minority class instances and Tomek links to clean the borderline majority class samples, this study evaluates the effectiveness of this hybrid method in improving classification metrics. Using a benchmark credit card fraud dataset, we compare the performance of Random Forest models with and without the hybrid sampling approach. The experimental results show that SMOTE-Tomek significantly enhances recall and F1-score without sacrificing accuracy. This finding underscores the importance of using appropriate resampling strategies for improving model robustness in fraud detection.