81.6% recall | 87.4% F1 | 284k transactions
Random Forest model that detects credit card fraud. End-to-end pipeline from data to deployment.
| Metric | Score |
|---|---|
| Recall | 81.6% |
| F1 | 87.4% |
- Engineered 7 features (transaction hour, user velocity, amount patterns)
- Trained a Random Forest model with 81.6% fraud catch rate
- Pushed predictions to SQL Server with complex queries (window functions, CTEs)
- Saved model with versioning for production
- Built a PyTorch neural network (11 parameters) to show framework knowledge
- Flag transactions > $106
- Monitor 2-3 AM (peak fraud hour)
- Approve users with >50 daily transactions faster (0.17% fraud rate)
Python · pandas · scikit-learn · PyTorch · SQL Server · joblib
Fixed random seed (42) for exact replication