Bagging or Boostingƒ Anomalous Transaction Identification in Financial Systems
Srushti Patel, Ruobin Qi, Rashida Hasan · 2025
Financial fraud detection has become a major concern in today’s financial infrastructure, where anomalous transactions often signal fraudulent activities. Detecting these anomalies is complicated by the highly imbalanced nature of transaction datasets. In this paper, we provide a systematic evaluation of ensemble methods for anomalous transaction detection across four diverse datasets: Credit Card Fraud, PaySim, Insurance Fraud, and Car Loan Defaulter. We focus on two categories of ensemble methods: Bagging (Random Forest, Extra Trees, and Isolation Forest) and Boosting (AdaBoost, Gradient Boosting, and XGBoost). We compare these models based on multiple criteria, such as accuracy, F1 score, precision, and recall, to compare their effectiveness.. Experimental results indicate that Boosting methods, particularly XGBoost, Gradient Boosting, and AdaBoost, outperform Bagging methods in classification performance across most datasets. However, Bagging methods such as Random Forest and Extra Trees remain competitive in datasets like Insurance Fraud. While Boosting methods typically excel in detecting fraudulent transactions, Bagging methods show strong performance in certain instances. This paper contributes to a deeper understanding of how ensemble methods can enhance the reliability and effectiveness of detecting fraudulent activities in diverse financial settings.