NCAA DIII Football Prediction ModelJuly 2025 — present
An end-to-end system that predicts Division III football outcomes, averaging 79.95% accuracy on game winners through the 2025 season. It pulls game data through a GraphQL ETL pipeline, derives 70+ features including rolling team metrics and a custom Elo rating, and trains calibrated logistic regression models with isotonic regression and time-based cross-validation. The Elo system alone improved accuracy by more than 15%.
Python · pandas · scikit-learn · PostgreSQL · GraphQL · SQLAlchemy