WNBA Player Scoring Predictor
My first real swing at machine learning, pointed at a sport that deserves more data love
A random forest model trained on nearly 3,900 WNBA player-game records (28 features, sourced from FiveThirtyEight's WNBA stats archive) to predict a player's true shooting percentage — tuned via GridSearchCV to an R² of 0.87 on the test set. Points per game, a composite performance rating, and shooting efficiency turned out to be the strongest predictors, well ahead of usage rate or minutes played. First real dive into feature engineering and model tuning, and a chance to put some data rigor behind a sport that deserves more of it. Next: turning it into an interactive app where you pick a player and get a live prediction.
Python, scikit-learn · March 2025
Tested on data the model had never seen — an R² of 0.87, strong evidence it learned real scoring patterns rather than memorizing the training set.
Built from nearly 3,900 player-game records — shooting efficiency, pace, usage rate — engineered into the features that actually predict scoring, not just logged for the record.