Higher Precision, Faster Deployment: Advancing Beyond the Wco Bacuda Lite Date with an Optimized Fraud Detection Model

This report presents the development of an optimized XGBoost model for detecting illicit customs declarations, benchmarking its performance directly against the World Customs Organization's (WCO) LITE DATE model. While the WCO's model established a vital and accessible benchmark (Recall 79.9%, Precision 20.8%), our objective was to enhance its operational efficiency by reducing false positives without sacrificing detection rates. Through advanced feature engineering and hyperparameter tuning, our resulting model demonstrates a significant performance lift. We achieved a superior F1-Score (0.35 vs. 0.33) and, critically, boosted precision to 22.0% while maintaining a comparable recall of 79.0%. This represents a more intelligent model that provides higher-quality alerts, measurably reducing analyst workload and amplifying the strategic value of the detection system. We conclude that this optimized model is the next evolution in accessible customs analytics and is ready for operational deployment.