Author: Romeo Thomas
Org: Contranemic
Category: Machine Learning / Financial Risk Analytics
Language: Python 3.11+
Fraud ML Suite is a production-ready pipeline for training, evaluating, and deploying fraud detection models using gradient-boosted ensembles with explainability (SHAP).
It includes a FastAPI scoring service, Docker deployment workflow, and CI testing.
- Gradient Boosting (LightGBM) with Bayesian hyperparameter tuning
- Feature store with rolling-window aggregations
- Class imbalance handling (SMOTE + cost-sensitive objective)
- XAI visuals via SHAP and Plotly
- FastAPI inference API + GitHub Actions CI
git clone https://github.com/Contranemic/fraud-ml.git
cd fraud-ml
pip install -e .
pytest
uvicorn api.main:app --reload