Total Features
125
in production model
Training Games
34,694
2022-2026 seasons
SHAP Sample
5,000
random games analyzed
Top Feature
srs_diff
mean |SHAP| = 0.463
Explainer Type
Tree
exact (not approximate)
Model Accuracy
68.8%
validated on held-out data
Beeswarm Plot (Top 40)
Bar Chart (All 125)
Full Table
SHAP Beeswarm — Top 40 Features
Each dot = one game. X-axis = SHAP value: positive pushes toward home win, negative toward away win. Dot color = feature value (red=high, blue=low). Hover a row for details.
Feature value:
Low
Mid-low
Mid
Mid-high
High
About SHAP: SHAP (SHapley Additive exPlanations) values measure each feature's marginal contribution to a prediction.
A SHAP value of +0.10 means that feature pushed the predicted home-win probability up by ~10 percentage points for that game.
TreeExplainer computes exact (not approximate) SHAP values for XGBoost models.
Sample: 5,000 games randomly drawn from 33,462 training games (2022-2026 regular season). Target variable: home_won (1 = home team won, 0 = away team won).
Sample: 5,000 games randomly drawn from 33,462 training games (2022-2026 regular season). Target variable: home_won (1 = home team won, 0 = away team won).