Enhancing Customs Fraud Detection: A Comparative Study of Methods for Performance Measurement and Feature Improvement

This article presents a comprehensive investigation into the augmentation of customs fraud detection capabilities through the integration of Random Forests, XGBoost, and Ensemble Stacking techniques. In response to the growing intricacies of fraudulent activities in cross-border trade, this study explores advanced machine learning methods as vital solutions. Leveraging a meticulously preprocessed dataset of customs import data, Random Forests and XGBoost models were extensively trained to pinpoint fraudulent transactions with precision.Performance metrics including accuracy, precision, recall, and F1-score were systematically employed to evaluate the models. Moreover, graphical representations, encompassing confusion matrices, feature importance plots, and log-loss curves, provided insightful visualizations for performance comparison. The analysis highlights the distinctive strengths and weaknesses of Random Forests and XGBoost models in customs fraud detection, further accentuated through Ensemble Stacking. This work not only emphasizes their competence in flagging fraudulent transactions but also elucidates the hierarchical importance of individual features.The results obtained pave the way for fortified fraud detection systems, reinforcing the security and integrity of cross-border trade operations in response to evolving threats.