WPA-POWERED BASKETBALL ANALYTICS

Who contributes to winning — at every level.

Possession-level win probability impact for every player, pair, and lineup. Validated with r=0.22 cross-season prediction.

—

Players

—

Pairs scored

—

Possessions

0.22

Prediction r

Who wins together

Win contribution per 100 possessions, EB-shrunk

Shapley WPA per 100 possessions

5-man groups by win contribution

Win probability model

LSTM trained on play-by-play data predicts home win probability at every event. WPA = change per possession.

Shapley attribution

Ridge regression assigns WPA to players (coefficients = Shapley values). Cross-season r=0.22.

Every grouping level

Score players, pairs, trios, quartets, lineups. Empirical Bayes shrinkage. Confidence tiers.