Prediction Models
The models that predict every game — what they know, and how accurate they really are against the sportsbook closing line.
Model 4
DefaultMachine-learned model served from a trained bundle.
- Spread MAE
- 8.87 (book 8.83)
- Log loss
- 0.5095 (book 0.5251)
Machine-learned model served from a trained bundle.
- Spread MAE
- 9.11 (book 8.83)
- Log loss
- 0.5280 (book 0.5251)
Machine-learned model served from a trained bundle.
- Spread MAE
- 8.96 (book 8.83)
- Log loss
- 0.5221 (book 0.5251)
Machine-learned model served from a trained bundle.
- Spread MAE
- 9.14 (book 8.83)
- Log loss
- 0.5505 (book 0.5251)
Classical model computed from adjusted per-possession efficiency ratings.
- Spread MAE
- 9.39 (book 8.83)
- Log loss
- 0.5629 (book 0.5251)
All-season aggregates mix seasons some ML models trained on (in-sample) with held-out ones — see each model's About page for the honest per-season breakdown. The book figures are the sportsbook closing line on the same games. Not betting advice.