archived

See the game before it's played.

NFL Nostradamus was a full-stack machine learning application that predicted NFL game winners and point spreads using 20 seasons of play-by-play data. The classifier achieved 62.89% validation accuracy on the 2023 season — beating both home-team and better-record baselines — using a temporally split XGBoost pipeline that eliminated the data leakage common in hobby ML projects. A spread regression model extended predictions beyond win/loss into point differential territory, all served through a FastAPI backend, persisted in PostgreSQL, and rendered in a React dashboard. The public app has been retired; the historical data and source code are preserved.