------------------------------------| 1. DATA SOURCES |
| Kaggle Credit Card Fraud, |
| open banking transaction logs, |
| real-time payment streams, and |
| SWIFT-like synthetic generators |
| provide high-volume financial |
| signals. Data is labeled for |
| fraud patterns and inspected for |
| temporal irregularities common |
| in online payment flows. |
------------------------------------ | v ----------------------------------- | 2. INGESTION & PREPROCESSING |
| Streaming pipelines validate |
| transaction schemas, dedupe |
| repeated entries, and enforce |
| chronological ordering. |
| Numerical fields are normalized |
| and geographic metadata is |
| enriched. |
----------------------------------- | v ------------------------------------ | 3. FEATURE ENGINEERING |
| Time-based movement windows, |
| device fingerprints, merchant |
| embeddings, and account relation |
| graphs are constructed. |
| Behavioral transitions are |
| encoded. |
------------------------------------ | v ---------------------------------- | 4. MODEL TRAINING |
| (Autoencoder + Isolation |
| Forest) Autoencoders learn |
| normal patterns. Isolation |
| Forest captures rare behaviors.|
| Tuning improves detection. |
---------------------------------- | v --------------------------------- | 5. INFERENCE & ALERTING |
| Low-latency scoring engine |
| evaluates transactions. High- |
| risk trigger alerts. Batch |
| scoring supports auditing. |
--------------------------------- | v --------------------------------- | 6. INVESTIGATION & |
| FEEDBACK LOOP |
| Fraud analysts review cases. |
| Confirmed enrich training. |
| Monitoring tracks drift. |
---------------------------------