Interactive overview of fraud patterns, model performance, and risk segmentation — 11,903 transactions analysed
How fraud distributes across transaction types, time, and flag combinations
| No Gift Card | Gift Card | |
|---|---|---|
| Domestic | 5.2% | 36.8% |
| International | 45.3% | 89.3% |
Browsing patterns that distinguish fraudulent sessions from legitimate ones
Four models trained with temporal validation and hyperparameter tuning
| Model | AUC | Strength |
|---|---|---|
| Gradient Boosting | ~0.95 | Best overall |
| Random Forest | ~0.93 | Robust ensemble |
| Logistic Regression | ~0.92 | Interpretable |
| Decision Tree | ~0.82 | Readable rules |
Data-driven thresholds optimised for maximum fraud capture with minimum review burden