Inside the CHD Scout prediction engine powering our game forecasts
Model Overview
The CHD Scout prediction model is a multi-factor system that analyzes team efficiency, player form, venue effects, and schedule context to forecast the outcome of every NCAA Division I basketball game. For each matchup, the model produces a predicted final score, win probability, and confidence rating.
The model runs entirely in SQL within our Supabase database, ensuring predictions are consistent and reproducible. Every prediction is recorded before tipoff and graded after the game, creating a fully transparent track record.
Key Inputs
The model evaluates the following factors for each game:
1. Offensive & Defensive Efficiency: Points per 100 possessions, adjusted for opponent quality. This is the single most predictive metric in college basketball analytics.
2. NET Rankings: The official NCAA ranking that captures overall team quality, strength of schedule, and body of work.
3. Player Form: Our 5-game weighted PRA (Points + Rebounds + Assists) model tracks whether key players are trending up or down.
4. Venue-Specific Home Court Advantage: Not all home courts are equal. Our model scales HCA based on venue strength rather than using a flat national average.
5. Rest & Travel: Teams playing back-to-back or traveling long distances receive adjustments.
6. Conference Familiarity: Rematches within the same conference are adjusted for the familiarity factor.
7. Star Player Absence: When a team's leading scorer (14+ PPG) is unavailable, the model adjusts expectations.
Prediction Architecture
The model uses a nudge-based architecture with multiple stages:
Stage 1 - Base Margin: Calculated from efficiency differentials between the two teams, with adjustments for opponent quality.
Stage 2 - Home Court Advantage: Applied based on venue, with a global average of 3.5 points and a ceiling of 6.0 points for the strongest venues.
Stage 3 - Competitive Boost: In projected close games (predicted margin under 9 points), an additional boost amplifies the home court effect, reflecting the observed tendency of home teams to outperform in tight games.
Stage 4 - Nudge Model: For games with moderate or toss-up margins, a secondary model blends efficiency, NET ranking, and recent form signals to refine the prediction. This prevents the model from being overconfident in true toss-up games.
Stage 5 - Contextual Adjustments: Rest days, conference familiarity, and star absence modifiers are applied.
Accuracy Track Record
Our model's accuracy is tracked in real-time on our Accuracy page. Key metrics for the 2025-26 season:
- Overall Winner Accuracy: 76%+ across 4,000+ games
- Mean Absolute Error: ~9.3 points (actual vs predicted score)
- High-Confidence Games (top quartile): 85%+ accuracy
- Toss-up Games (margin < 3): ~55% (inherently unpredictable)
The model is strongest for games with clear favorites (large efficiency gaps) and weakest for true toss-ups where small factors like in-game momentum swings dominate.
How We Compare to Other Sites
College Hoops Data vs KenPom ($24.95/year):
- Both use efficiency-based metrics as the foundation
- CHD is completely free with no paywall
- CHD adds player form tracking, AI-generated articles, and an interactive tournament simulator
- CHD provides transparent, real-time accuracy tracking
College Hoops Data vs BartTorvik (free):
- Both offer free efficiency data
- CHD adds game predictions, player hot/cold analysis, and AI articles
- CHD has a more polished, mobile-friendly interface
College Hoops Data vs ESPN BPI:
- ESPN BPI is embedded within ESPN's ecosystem
- CHD offers more detailed matchup analysis and transparent methodology
- CHD provides quad record analysis and tournament resume grading
Frequently Asked Questions
How accurate is the College Hoops Data prediction model?
The CHD Scout model has achieved 76%+ winner accuracy across 4,000+ games in the 2025-26 season, with a mean absolute error of approximately 9.3 points. For high-confidence games (top quartile by predicted margin), accuracy exceeds 85%. True toss-up games (predicted margin under 3 points) sit around 55%, which is expected since those games are inherently unpredictable.
Is College Hoops Data free to use?
Yes. College Hoops Data provides all predictions, NET rankings, quad records, player stats, and analytics completely free with no paywall. Unlike KenPom ($24.95/year), every feature on CHD is accessible without a subscription.
How does the CHD prediction model compare to KenPom and BartTorvik?
Like KenPom and BartTorvik, CHD uses adjusted efficiency metrics as its foundation. CHD adds several layers on top: player form tracking (hot/cold analysis), venue-specific home court advantage scaling, competitive game boosts, rest/travel adjustments, and conference familiarity dampening. CHD also provides transparent real-time accuracy tracking so you can verify the model's performance yourself.
When are predictions published for each game?
Predictions are generated and recorded before tipoff for every Division I game. They are captured when the game status changes to live, ensuring no look-ahead bias. Every prediction includes the model version, predicted scores, win probability, and confidence rating, all of which are permanently stored for accuracy grading.