Model Card — Fraud Pattern Detector
Model details
- Task: binary classification (fraud vs non-fraud)
- Baseline model: XGBoost with compact feature set + frequency encoding for selected categorical fields
- Explainability: SHAP TreeExplainer (top driver contributions)
Intended use
- Educational/demo fraud scoring and pattern exploration.
- Not for production blocking decisions without proper monitoring, retraining, and policy controls.
Training data
- Source: Kaggle “IEEE-CIS Fraud Detection” competition dataset (user must accept Kaggle rules).
- Target:
isFraud
Metrics (holdout split)
See artifacts/metrics.json.
Limitations
- Uses a simplified feature set to keep the demo light and portable.
- The “network patterns” tab is illustrative of cluster behavior; it is not the full dataset graph.