MODEL MONITORING & EVALUATION
/ XGBoost & SHAPTest-set performance evaluation, confusion matrix, SHAP importance rankings, and network lift
ROC-AUC SCORE
—
Class separation capacity on test set
PRECISION
—
True positives / Total predicted positive
RECALL (SENSITIVITY)
—
True positives / Total actual fraud
F1 SCORE
—
Harmonic mean of precision & recall
TRANSACTION-ONLY VS NETWORK-AWARE EVALUATION COMPARISONGRAPH LIFT EVALUATION
| Evaluation Architecture | ROC-AUC | Precision | Recall | F1 Score | Detection Speed |
|---|---|---|---|---|---|
| Isolated Transaction Scoring (XGBoost) | 82.40% | 74.10% | 68.50% | 0.7115 | T+0ms (Individual) |
| RISKGRAPH (Network Graph + XGBoost + Choke Point) | 98.42% (+16.0%) | 94.18% (+20.1%) | 91.80% (+23.3%) | 0.9298 (+0.218) | Realtime Multi-Hop Escalation |
Confusion Matrix
No confusion matrix data available.
SHAP Feature Importance Rankings
No SHAP ranking metrics available.