Abstract—As the volume, structural complexity, and diversity of participants in international trade continue to expand, customs administrations face increasing pressure to accurately identify high-risk import declarations under strictly limited inspection resources. Although machine learning-based approaches to fraud and anomaly detection in customs data have proliferated in recent years, most existing studies primarily emphasise classification performance while largely overlooking practical inspection capacity constraints, asymmetric economic costs of misclassification, and the ability of models to generalize to future, previously unseen declarations. This study analyses a large-scale dataset comprising 628,015 construction material import transactions recorded between January 1, 2023, and September 30, 2025. We develop and empirically evaluate a risk-scoring framework that integrates time-aware evaluation, cost-sensitive decision logic accounting for asymmetric misclassification costs, and explainable artificial intelligence techniques based on Shapley Additive Explanations (SHAP). Rather than framing customs risk assessment as a binary classification task, the proposed approach prioritizes import declarations through risk-based ranking that explicitly reflects real-world inspection capacity constraints. Empirical results demonstrate that risk-score-based ranking enables more efficient allocation of limited inspection resources than conventional classification-based approaches. Time-aware evaluation provides a more realistic assessment of model generalization to future declarations, while cost-sensitive decision logic aligns risk assessment with its underlying economic consequences. SHAP-based explanations further reveal that the behavioral characteristics of import declarations primarily drive risk scores. Overall, the findings provide empirical evidence that effective deployment of machine learning in customs risk management requires the joint consideration of inspection capacity constraints, asymmetric economic costs, and institutional interpretability.
 
Keywords—customs risk management, machine learning, trade mis-invoicing, import declaration analysis, cost-sensitive decision making
 
Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).