Mutual Funds Predictor

An end-to-end forecasting system for Indian mutual funds. Every day a GitHub Actions pipeline ingests fresh NAVs for about 20,000 funds, rebuilds risk features and re-scores them with a stacked ensemble. Explanations and drift monitoring sit in a Streamlit app.

Approach

1. Stacked ensemble: XGBoost, Random Forest and Ridge, blended by a meta-learner.
2. Walk-forward validation, so the model is always tested on a future it has not seen.
3. Daily MLOps pipeline on GitHub Actions ingesting NAVs for about 20,000 funds.
4. Rolling Alpha, Beta and Sharpe features computed per fund.
5. SHAP explainability and data-drift monitoring surfaced in Streamlit.