03 – Anomaly Detection (Fraud Prevention)

Synthetic fraud/anomaly detection pipeline that trains Isolation Forest + LOF on 50k hourly transactions, tracks runs in MLflow, exposes Streamlit + FastAPI for dashboards/RPA, and outputs alerts.

Setup

Set directory:

cd /Users/andresverafigueroa/Documents/GitHub/Works/03-anomaly-detection-fraud-prevention

1. Create venv + install deps

   python3 -m venv .venv && source .venv/bin/activate
   pip install -r requirements.txt
   

2. Point MLflow locally

   export MLFLOW_TRACKING_URI="$(pwd)/mlruns"
   

Quickstart

1. Train model + log artifacts

   python3 scripts/train.py --config configs/config.yaml
   

   Outputs: model joblib, metrics JSON, scores CSV, MLflow run.
2. Serve FastAPI for external callers

   ./scripts/run_api.sh
   

   Endpoints: /health, /score (single transaction), /batch_score, /metrics.
3. Start Streamlit dashboard

   streamlit run src/app.py -- --config configs/config.yaml
   

4. Use Power BI / other dashboards
   • Point to http://localhost:8000/score or /batch_score with JSON payload for live anomaly data.

Repo Layout

• configs/ – config (paths, model params, threshold, alerts)
• data/03-anomaly/ – raw + processed CSVs
• models/03-anomaly/ – serialized pipeline
• artifacts/ – metrics, scores, alerts
• src/ – data gen, features, model training, API, dashboard
• scripts/ – train, run API, serve Streamlit, run alerts
• tests/ – unit + integration tests
• docs/ – reference notes

Key Commands

• Train: python3 scripts/train.py --config configs/config.yaml
• API: ./scripts/run_api.sh
• Dashboard: streamlit run src/app.py -- --config configs/config.yaml
• Tests: pytest -q

Notes

• Alert rules write alerts.jsonl with flagged transactions; inspect it to feed RPA.
• FastAPI payload must match configs schema (see src/api/app.py).