Spaces:
Running
Running
| title: Real-Time Fraud Detection | |
| emoji: 🛡️ | |
| colorFrom: green | |
| colorTo: blue | |
| sdk: gradio | |
| sdk_version: 5.50.0 | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| # 🛡️ Real-Time Credit Card Fraud Detection | |
| End-to-end fraud detection on the Sparkov dataset (1.85M transactions, ~0.5% fraud): | |
| - **Feature engineering** — leakage-safe velocity, behavioral, and geo features | |
| - **Imbalance study** — cost-sensitive vs SMOTE vs undersampling (honest comparison) | |
| - **Three models** — LightGBM (PR-AUC 0.97) · GraphSAGE GNN · Autoencoder | |
| - **SHAP explainability** — per-transaction reasons for compliance | |
| - **Concept-drift monitoring** — PSI early warning | |
| - **Real-time scoring** — single-digit-millisecond latency | |
| Source: [github.com/Fikri645/fraud-detection](https://github.com/Fikri645/fraud-detection) | |
| Built by [Muhammad Fikri Wahidin](https://github.com/Fikri645). | |