This study presents a credit card fraud detection framework integrating feature engineering, optional SMOTE, XGBoost, MCC--cost threshold optimization, and SHAP explainability. After duplicate removal, European transaction data were split into stratified training, validation, and test sets. Data-dependent preprocessing and SMOTE used training data only; threshold selection used validation data. Independent test performance reached ROC-AUC 0.9698, PR-AUC 0.8136, and MCC 0.8319, with a classification cost of 228. Evaluation included bootstrap intervals, McNemar testing, ablation, cost sensitivity, and PSI analysis. Operational outputs are APPROVE, REVIEW, and BLOCK. A 1,000-sample batch benchmark supports deployment feasibility, subject to infrastructure-dependent production latency. Low PSI indicates train--test distributional similarity; long-term temporal stability and adaptation remain untested.
Kaveh, N. (2026). A System-Oriented Framework for Credit Card Fraud Detection Using Feature Engineering and Threshold Optimization. Journal of Data Science and Modeling, (), 1–30. https://doi.org/10.22054/jdsm.2026.93423.1095
MLA
Kaveh, N. "A System-Oriented Framework for Credit Card Fraud Detection Using Feature Engineering and Threshold Optimization", Journal of Data Science and Modeling, , 2026, 1–30. doi: 10.22054/jdsm.2026.93423.1095
HARVARD
Kaveh, N. (2026). 'A System-Oriented Framework for Credit Card Fraud Detection Using Feature Engineering and Threshold Optimization', Journal of Data Science and Modeling, (), pp. 1–30. doi: 10.22054/jdsm.2026.93423.1095
CHICAGO
Kaveh, N., "A System-Oriented Framework for Credit Card Fraud Detection Using Feature Engineering and Threshold Optimization." Journal of Data Science and Modeling, (2026): 1–30, doi: 10.22054/jdsm.2026.93423.1095
VANCOUVER
Kaveh N. A System-Oriented Framework for Credit Card Fraud Detection Using Feature Engineering and Threshold Optimization. JDSM. 2026;():1–30. doi: 10.22054/jdsm.2026.93423.1095
