Probabilistic Yard Gain Forecasting
Calibrated play-outcome modeling from NFL tracking data, framed as a full cumulative distribution rather than a point estimate.
Overview
Framed yard-gain prediction as a probabilistic problem and predicted the complete cumulative distribution of outcomes rather than a single point estimate.
Selected results
- Engineered play-level spatial and football-specific features from tracking data.
- Benchmarked logistic regression, SVM, random forest, gradient boosted trees, and XGBoost on held-out data.
- Calibrated forecast probabilities with isotonic regression for threshold-based decisions.
- Served the model through a Flask application for interactive evaluation.