Forecasting NBA Player Fantasy Points
A predictive modeling framework for forecasting season-average NBA player statistics using gradient boosting, allowing fantasy projections under any custom scoring system.
Commercial fantasy sports products forecast per-game statistics, which are highly volatile and prone to noise, making long-term roster construction and trade evaluation unreliable.
The team trained Gradient Boosting regressors on multi-season historical NBA box-score data to predict season-average totals across 12 statistical categories plus games played, then computed fantasy point projections across custom league scoring rules (e.g. ESPN, Yahoo, FanDuel).
Showed significant reduction in prediction error compared to per-game forecasting baselines, serving both fantasy projections and front-office trade/draft evaluation.