Real-Time Fraud Detection
Sub-second authorization fraud scoring
owner: Fraud Data Science · last validated Apr 18, 2026
Outcome it's accountable for
metric: fraud $ caught vs rules baseline
Top feature drivers (SHAP)
Disaggregated performance
| segment | AUC | approval |
|---|---|---|
| Mass | 0.939 | 97.1% |
| Affluent | 0.945 | 97.8% |
| Private Client | 0.948 | 98.2% |
| Small Business | 0.933 | 96.4% |
Tested for disparate impact across segments - no material divergence.
Intended use
Score each authorization in real time for fraud probability; decline / step-up above threshold. Decision-support with human review on edge cases.
Training data
18 months of authorizations (bronze) + confirmed-fraud labels. Class imbalance handled with SMOTE; evaluated on a held-out temporal split.
Limitations
Performance degrades on novel CNP attack patterns; retraining triggered on KS drift > 0.05. Not used for credit decisions.
Fair lending & adverse action
No protected-class attributes used. Disparate-impact tested across region and segment; SHAP reason codes generated for every decline (adverse-action support).