Nightly NBA predictions — with receipts.
Three machine-learning models pick every NBA game. Every prediction is logged, scored against the real result, and published — hits and misses alike.
Predictions for 2026-10-30
| Matchup | Models' pick | Neural Net | Random Forest | XGBoost | Market |
|---|---|---|---|---|---|
| CHA @ WAS WAS 95–79 | WAS 52%✓coin flip 2 of 3 models agree | CHA by 1.2 CHA 56% | WAS by 1.9 WAS 53% | CHA by 1.2 WAS 58%split | CHA -1.5 |
| CHI @ BOS BOS 125–111 | BOS 78%✓ 3 of 3 models agree | BOS by 10.1 BOS 82% | BOS by 4.8 BOS 71% | BOS by 8.9 BOS 82% | BOS -15.5 |
| DET @ BKN DET 110–94 | DET 52%✓coin flip 2 of 3 models agree | BKN by 1.1 DET 50%split | BKN by 1.5 DET 55%split | BKN by 0.1 BKN 51% | DET -7.5 |
| NYK @ PHI PHI 98–93 | PHI 55%✓coin flip 2 of 3 models agree | PHI by 1.1 NYK 52%split | PHI by 1.9 PHI 52% | PHI by 4.0 PHI 66% | NYK -1.5 |
| ORL @ TOR ORL 121–114 | TOR 67%✗ 3 of 3 models agree | TOR by 5.8 TOR 69% | TOR by 4.1 TOR 64% | TOR by 8.5 TOR 67% | TOR -5.5 |
| IND @ MIA | MIA 64% 3 of 3 models agree | MIA by 8.5 MIA 68% | MIA by 3.0 MIA 58% | MIA by 1.5 MIA 66% | MIA -7.5 |
| HOU @ DAL | HOU 54%coin flip 3 of 3 models agree | HOU by 1.3 HOU 52% | DAL by 1.2 HOU 55%split | DAL by 1.0 HOU 57%split | HOU -5.5 |
| LAC @ MIN MIN 127–101 | MIN 54%✓coin flip 2 of 3 models agree | MIN by 1.8 MIN 53% | MIN by 3.0 MIN 59% | MIN by 4.4 LAC 49%split | MIN -7.5 |
| LAL @ GSW | GSW 50%coin flip 2 of 3 models agree | GSW by 3.2 GSW 53% | GSW by 2.5 GSW 59% | GSW by 0.3 LAL 61%split | LAL -1.5 |
| DEN @ PHX | PHX 54%coin flip 3 of 3 models agree | PHX by 2.3 PHX 54% | PHX by 2.3 PHX 57% | DEN by 0.8 PHX 51%split | — |
One row per game. "Models' pick" is the team favored by the average of the three models' win probabilities, and how many models agree. Each model column is the margin that model expects (green if that team won, red if not), with its own win probability underneath. Margin and win probability come from two separately trained parts of each model, so they can occasionally favor different teams — flagged "split". "Market" is the betting line ("BOS -6.5" means BOS favored by 6.5), shown as a benchmark, not a pick. Hover a cell for details.
How it works
Box scores in
Every game's team stats — shooting, rebounding, hustle, tracking.
Rolling form
Each team's recent play, distilled into rolling features.
Three models
A random forest, gradient boosting and a neural net each pick a winner and a margin.
Want the longer version — data, models, the nightly cycle? How HoopModel works →