Year: 2026 | Month: June | Volume: 13 | Issue: 6 | Pages: 219-235
DOI: https://doi.org/10.52403/ijrr.20260622
SMOTE-Based Supervised Learning Approaches for Import Cargo Routing Classification in Indonesian Customs
Affan Rafi Ardiansyah1, Prajna Pramita Izati2
1Department of Accounting, State Financial Polytechnic STAN, Tangerang Selatan, Indonesia
2Department of Statistics, Faculty of Science and Mathematics, Diponegoro University, Semarang, Indonesia
Corresponding Author: Prajna Pramita Izati
ABSTRACT
This study compares the performance of several supervised learning methods in determining import lane status at KPPBC TMP B Teluk Bayur using variables such as CIF value, country of origin, importer status information, and HS Code. The evaluated methods include Logistic Regression, Support Vector Machine (SVM), Naive Bayes, and Extreme Gradient Boosting (XGBoost), with SMOTE applied to address imbalanced data. The results show that all models are capable of classifying import documents into green lane and red lane categories. Among the models, XGBoost achieved the best overall performance with an accuracy of 96.3% and a balanced precision and recall value. Logistic Regression and SVM showed high recall values of 88.9%, indicating strong capability in detecting red lane cases, while Naive Bayes demonstrated lower overall performance. SHAP analysis revealed that CIF value, country of origin, importer status information, and HS Code were the most influential variables in determining import lane status. Overall, XGBoost can be considered the best model, although model selection should still depend on operational priorities between efficiency and risk detection capability.
Keywords: supervised learning, import lane classification, XGBoost, SHAP, customs risk management, SMOTE
[PDF Full Text]