How do I build an NBA player prop model?
Start with one row per line: the line before the game, the tier, your features and the result, then split by date so the future never leaks into training. A model only has skill if it beats the base over rate for each market and tier out of sample, and the table below gives you those base rates.
| Market | Standard over rate | Standard rows | Goblin over rate | Demon over rate |
|---|---|---|---|---|
| Points | 43.9% | 726 | 64.1% | 20.0% |
| Pts+Rebs | 46.0% | 695 | 64.2% | 22.9% |
| Pts+Rebs+Asts | 44.4% | 723 | 62.5% | 21.4% |
| Rebounds | 47.9% | 422 | 69.9% | 19.3% |
| Pts+Asts | 46.2% | 563 | 63.6% | 21.9% |
| Rebs+Asts | 49.7% | 459 | 65.6% | 21.3% |
| Assists | 44.8% | 281 | 68.6% | 18.7% |
| 3-PT Made | 53.9% | 165 | 66.1% | 18.3% |
| Steals | 50.0% | 172 | 67.0% | 14.6% |
| Blks+Stls | 50.4% | 119 | 65.6% | 20.4% |
GET /api/v1/training-set gives labeled NBA player-prop rows in JSON or CSV: recorded lines and tiers alongside actual_value and result. Add prior-game form from /api/v1/game-logs and matchup features from /api/v1/team-stats and /api/v1/opponent-stats. Use only games completed before the prediction cutoff; actual_value, result and the predicted game’s minutes are labels or post-game observations, never pre-game features.
Split training, validation and test sets by game date, keeping every line for a game in the same split. Start with a simple baseline and evaluate calibration and error on later games. Line history includes a partial NBA capture on 27 Oct 2026, with broader recording from 28 Oct 2026; older loaded box scores do not create older PrizePicks lines. Free gives a sample of up to 100 training rows; Developer pages the full available set and reads full seasons of game logs. No model or dataset guarantees a profitable result.
Sample labeled NBA training rows
curl "https://prizepicksapi.com/api/v1/training-set?league=NBA&stat=Points&tier=standard&limit=100" \
-H "Authorization: Bearer $PRIZEPICKS_API_KEY"Go further
Related questions
Independent data service, not affiliated with PrizePicks. Information only, not betting advice.