Hoop Predict
R-based WNBA player scoring prediction model using rolling statistics, opponent defense, and home/away context.
R · RStudio · OLS · Ridge · Time-based validation
Hoop Predict asks how well you can forecast WNBA player scoring from recent form, opponent defense, and home/away context. I engineered those features in R, compared OLS and Ridge with time-based validation, and reached 4.18 MAE and 0.566 R² on held-out data.
Problem
Player scoring jumps around with form, matchup, and venue. Hoop Predict tries to forecast WNBA points using information you would actually have before a game.
The data
WNBA player and team game data covering recent seasons. Player statistics were aggregated into leakage-safe rolling 10-game averages for points, minutes, FGA, 3PA, FTA, rebounds, and assists. I also constructed a rolling 10-game opponent points-allowed metric and encoded home/away context. Rows without sufficient historical data were removed, and the data was split chronologically into training, validation, and test sets.
- Source
- WNBA game data
- Features
- Rolling player statistics · Opponent defense · Home/away
- Preprocessing
- Rolling averages · Missing-data removal · Time-based split
Feature engineering
The model uses rolling player statistics, opponent defensive metrics, and home/away context — signals that change with the matchup, not just a season average.
Models
I evaluated ordinary least squares and Ridge regression on the same feature set, to compare a simple linear baseline with a regularized alternative.
Time-based validation
I validated on held-out data using a time-based split, so the test set is later than the training set rather than a random mix of games.
Results
- MAE
- 4.18
- R²
- 0.566
On held-out test data, the model reached 4.18 MAE and 0.566 R².