Real-Time Fraud Detection Engine

flowchart LR
  A[Data generation] --> B[Feast feature store]
  B --> C[Training and MLflow]
  C --> D[FastAPI scoring]
  D --> E[Evidently drift monitoring]
  E --> F[Alert and retrain trigger]

Loading

How to run

1. pip install -r requirements.txt
2. python -m src.data_generator
3. python -m features.materialize_features, then from features/: feast apply and feast materialize-incremental <UTC ISO timestamp>.
4. python -m src.train
5. uvicorn api.main:app --reload --port 8000
6. python -m monitoring.drift_check
7. Follow LOAD_TESTING.md for Locust.

Results

src.train logs PR-AUC, precision, recall, and F1 for both models. Use Locust CSV output for throughput and p95 latency, and inspect the generated drift HTML reports under reports/. All reported numbers must come from an actual run on the deployment environment.