A Hybrid Oversampling Approach for Fraud Detection: Integrating SMOTE-ENN and ADASYN

Detecting financial fraud is challenging due to class imbalance in transactional datasets, where legitimate transactions vastly outnumber fraudulent ones.This imbalance biases traditional machine learning models toward the majority class, leading to high false negative rates despite high overall accuracy.To address this, the study proposes a hybrid oversampling method combining SMOTE-ENN and ADASYN to enhance detection performance.Initially, seven machine learning models were evaluated using SMOTE, with Random Forest, KNN, and XGBoost achieving the highest scores in accuracy, recall, and F1-score.These models were further tested using the proposed hybrid method, which integrates noise removal (via SMOTE-ENN) with adaptive minority sampling (via ADASYN).The hybrid approach significantly improved recall and F1-score, especially for Random Forest and XGBoost, achieving up to 99.99% accuracy.Results confirm that combining hybrid oversampling with robust classifiers reduces false negatives and improves generalization in fraud detection systems.