🐻⬇️🏀

2025-2026 NCAAMD2 Model Performance Analysis

Scope
All Games Ranked Games

All scored games in the selected league and season. AP Poll is excluded here.

Season
2025-2026

Comparing prediction accuracy across 1762 games using multiple rating models.

Model Catalog

7-day holdout coverage: 16/17 models .

Rolling Holdout Curves

Each point is a strict weekly holdout: train on all games before that week, test on that week. This first version uses a 21-day warmup, then 7-day holdouts stepped forward weekly.

Log Loss Brier AUC Accuracy

Weekly strict holdout log loss. Lower is better. Showing 16 models across 14 windows. Click legend items to hide/show series.

Recent Window Winners

Holdout Best Log Loss Runner-up Models
Feb 4 - Feb 7 Points Off/Def 0.528 Points Off/Def Recency (0.529) 16
Jan 28 - Feb 3 Core Ensemble 0.570 Recency Ensemble (0.570) 16
Jan 21 - Jan 27 Margin Recency 0.533 Margin (0.534) 16
Jan 14 - Jan 20 Margin 0.551 Points Off/Def (0.557) 16
Jan 7 - Jan 13 Margin 0.569 Points Off/Def (0.574) 16
Dec 31 - Jan 6 Margin 0.549 Core Ensemble (0.552) 16
Dec 24 - Dec 30 Margin 0.237 Points Off/Def (0.240) 16
Dec 17 - Dec 23 Margin 0.569 Points Off/Def (0.572) 16

Model Performance Leaderboard

Models ranked by strict holdout AUC when available (fallback: full-season AUC). Hover over column headers for explanations.

# Model 7d Split AUC Acc Brier LogLoss n AUC 7d Acc 7d Brier 7d n 7d
1 Points Off/Def Recency Points Off/Def Recency Off/def points regression with exponential recency weights. More → STRICT
285g
 - - - - 0 0.820 72.6% 0.178 285
2 Points Off/Def Points Off/Def Raw points regression with separate offensive and defensive team parameters. More → STRICT
285g
 0.796 72.9% 0.191 0.562 262 0.816 71.6% 0.178 285
3 Margin Margin Linear team-strength model fit on point differential instead of binary wins. More → STRICT
285g
 0.802 71.8% 0.184 0.544 262 0.813 73.0% 0.179 285
4 Margin Recency Margin Recency Margin regression with exponential recency weights on newer games. More → STRICT
285g
 - - - - 0 0.810 72.6% 0.180 285
5 Adjusted Context Blend Adjusted Context Blend Experimental context-heavy win model blending strong team components with rest and venue context. More → STRICT
285g
 - - - - 0 0.810 72.3% 0.180 285
6 Core Ensemble Core Ensemble Equal-logit blend of Elo, recency BT, recency margin, log-adjusted pyth, and points off/def. More → STRICT
285g
 - - - - 0 0.798 72.6% 0.183 285
7 Recency Ensemble Recency Ensemble Equal-logit blend of Elo, recency BT, recency margin, log-adjusted pyth, and recency points off/def. More → STRICT
285g
 - - - - 0 0.798 73.0% 0.183 285
8 Avg Margin Baseline Avg Margin Baseline Predict from simple average scoring margin in the training window. More → STRICT
285g
 0.874 77.9% 0.150 0.458 262 0.780 69.5% 0.191 285
9 Adjusted Efficiency Adjusted Efficiency Opponent-adjusted efficiency model with separate offensive and defensive components. More → STRICT
285g
 0.799 71.4% 0.184 0.541 262 0.779 73.0% 0.196 285
10 Log Adjusted Log Adjusted Log-scale adjusted efficiency model that downweights blowout leverage. More → STRICT
285g
 0.794 72.5% 0.187 0.548 262 0.779 73.3% 0.196 285
11 Bradley-Terry Bradley-Terry Static logistic paired-comparison model with one team strength parameter. More → STRICT
285g
 0.796 70.2% 0.185 0.547 262 0.777 69.8% 0.192 285
12 Dynamic Bradley-Terry Dynamic Bradley-Terry Time-evolving paired-comparison model with latent team strength drift. More → STRICT
285g
 - - - - 0 0.776 71.9% 0.192 285
13 Pythagorean Pythagorean Pythagorean win expectation from raw points scored and allowed. More → STRICT
285g
 0.787 70.6% 0.195 0.573 262 0.772 72.3% 0.199 285
14 Bradley-Terry Recency Bradley-Terry Recency Static Bradley-Terry with exponential recency weights on newer games. More → STRICT
285g
 - - - - 0 0.760 68.4% 0.198 285
15 Elo Elo Streaming paired-comparison rating with recency baked into sequential updates. More → STRICT
285g
 0.844 79.0% 0.179 0.540 262 0.760 70.2% 0.197 285
16 Efficiency Efficiency Tempo-adjusted efficiency version of Pythagorean ratings. More → FULL
no 7d
 0.638 57.6% 0.331 1.499 224 - - - 0
17 Home Team Baseline Home Team Baseline Always favor the home team with a fixed prior. More → STRICT
285g
 0.592 59.2% 0.242 0.676 262 0.635 63.2% 0.234 285

Methodology

ELO / Bradley-Terry

  • ELO: Iterative updates, K=64, HCA=100
  • BT: Static logistic regression on all games
  • Both model win probability, not margin
  • ELO updates after each game; BT fits once

Pythagorean Models

  • Raw: Classic points scored/allowed formula
  • Efficiency: Pace-adjusted (pts per possession)
  • Adjusted: Opponent-adjusted efficiency
  • Log: Log-linear multiplicative scale

Margin Regression

  • Team-level ridge regression on point margin
  • Linear Bradley-Terry (margin target)
  • Alpha=0.05 (CV-tuned)
  • Learns home advantage from data (~6 pts)

Baselines

  • Home Team: Always predict home wins (60%)
  • Avg Margin: Higher average margin wins
  • Models should beat these to add value