An Enhancement of AdaBoost Algorithm Applied In Online Transaction Fraud Detection System

This study is focused on the enhancement of the AdaBoost model for online transaction fraud detection to improve performance in detecting fraudulent activities. The study addresses the limitations of AdaBoost, including class imbalance, long training times, and overfitting. Three techniques were integrated to optimize the model. SMOTE (Synthetic Minority Over-sampling Technique) to balance the dataset, Quantile-Based Capping (QBC) to reduce training time, and Early Stopping to prevent overfitting. The dataset used for training consisted of online transaction records, with model performance evaluated using standard classification metrics. Results show that applying Heron-Centroid SMOTE led to an increase in accuracy, from 0.8150 to 0.8198, while significantly improving recall from 0.2962 to 0.3773, indicating better identification of minority class instances. The F-measure rose from 0.4135 to 0.4798, reflecting a better balance between precision and recall. The G-mean improved from 0.5337 to 0.5970, showing overall better classification performance. QBC reduced training time from 1.45 to 1.32 seconds, and Early Stopping increased accuracy from 0.73 to 0.816. These findings suggest that the proposed enhancements significantly improve AdaBoost’s efficiency and reliability, making it more effective for online fraud detection.