Explainable adaptive machine learning improves fraud detection in digital banking
Explainable and Adaptive Machine Learning for Fraud Risk Detection in Bangladesh's Digital Banking Ecosystem
Machine Learning
Summary
Detecting fraud in digital banking is hard because fraud is rare and patterns change over time. The authors created a machine learning system that not only identifies likely fraud but also explains which features influenced its decision, helping users understand why a transaction might be risky. Their system adapts the number of alerts to manageable levels, balancing false alarms with catching real fraud. They tested their approach on a simulated dataset reflecting mobile money transfers in Bangladesh, noting further work is needed to confirm results in real banking data.
What this means in practice
- •For fraud detection teams: Deploy adaptive alerting systems that prioritize high-risk transactions with explanations to improve fraud investigation efficiency.
- •For financial risk analysts: Use calibrated probability outputs and explainability tools to better interpret fraud risk scores in digital payment systems.
Tested on simulated data.
Authors
Md Najmus Swaqeeb
Abstract
The rapid expansion of digital banking and mobile financial services creates growing challenges for fraud detection, particularly under severe class imbalance and operational constraints. This study presents an explainable and adaptive machine-learning framework integrating chronological evaluation, leakage-conscious feature engineering, probability calibration, SHAP-based explainability, and adaptive risk prioritization. Experiments use the synthetic PaySim mobile-money benchmark and compare Logistic Regression, Random Forest, XGBoost, and LightGBM. XGBoost achieved the strongest fixed-threshold performance, with PR-AUC of 0.5658, ROC-AUC of 0.9068, and F1-score of 0.5369. Platt calibration reduced the test Brier score from 0.1141 to 0.0245, improving probability reliability. SHAP analysis identified temporal, destination-account, transaction-type, and amount features as important model drivers. An adaptive top-1% policy reduced false positives from 192 to 117, but recall declined from 0.4236 to 0.2122, demonstrating a clear alert-capacity trade-off. The results provide a reproducible methodological framework rather than evidence of deployment performance in Bangladesh, since PaySim is synthetic. Validation using governed real-world financial data is therefore an important direction for future research.