Explainable fraud detection under fixed false positive rate

Objective Fraud detection systems operate under extreme class imbalance and strict operational alert budgets, where false positives create substantial review cost and customer friction. This study aims to develop and evaluate an explainability-aware fraud detection pipeline that is explicitly assessed at an operationally constrained decision threshold (fixed false positive rate (FPR)) and to quantify how explanation patterns behave for true and false alerts in the high-precision regime. Methods We trained and compared four classifiers spanning interpretable baselines to high-performance ensembles: logistic regression, random forest, histogram-based gradient boosting, and Extreme Gradient Boosting (XGBoost), using the Credit Card Fraud Detection dataset (284,807 transactions; 492 frauds; 0.172% prevalence). To mirror real-world deployment, the decision threshold was tuned on a validation split to enforce FPR ≤ 1% and then applied unchanged to a held-out test set. Performance was reported using precision–recall area under the curve (PR-AUC), receiver operating characteristic area under the curve (ROC-AUC), recall at a fixed FPR, precision, and precision at K (precision@K). Statistical reliability was quantified via 100 stratified bootstrap resamples to obtain 95% confidence intervals. For interpretability, SHapley Additive exPlanations (SHAP) was used to compute global and local explanations, and we analyzed attribution concentration (fraction of total absolute SHAP mass captured by the top three features) for true positives (TPs) vs . false positives (FPs). Results At an FPR ≤ 1%, all models achieved strong fraud capture (recall ≈ 0.79–0.83). Random forest obtained the highest recall (0.827), whereas XGBoost achieved the highest PR-AUC (0.768). Logistic regression remained competitive and achieved the highest precision at the operating point (0.190). The precision at K (precision@100) was ~0.39–0.40 across all models, indicating comparable identification of the highest-confidence fraud cases. The explanation structure analysis showed substantial overlap between TPs and FPs, with mean attribution concentrations of ~54% and ~55%, respectively. Conclusion Strong detection performance can be achieved under operationally fixed alert budgets while providing decision-level explanations; however, SHAP attribution patterns for FPs can closely resemble those for true fraud in high-precision regimes. Explanations are valuable for transparency and investigator context but should be treated as decision support rather than reliable discriminators for triage.