Backtest performance
Walk-forward scoring: each fight is predicted using only prior UFC fights for both fighters, then ratings are updated with the result.
Scored fights
5,863
Brier
0.2444
Baseline Brier
0.2500
Log loss
0.6817
Accuracy
56.0%
Quick grid search
Best tested parameters: K low/high 36/14, temperature 1. Best grid log loss 0.6815, Brier 0.2443, accuracy 55.9%.
Reliability by predicted bucket
| Bucket | Fights | Predicted | Actual |
|---|---|---|---|
| 0.0%–10.0% | 0 | 0.0% | 0.0% |
| 10.0%–20.0% | 0 | 0.0% | 0.0% |
| 20.0%–30.0% | 4 | 28.5% | 75.0% |
| 30.0%–40.0% | 151 | 37.5% | 57.6% |
| 40.0%–50.0% | 1,882 | 46.3% | 59.3% |
| 50.0%–60.0% | 3,233 | 54.3% | 62.2% |
| 60.0%–70.0% | 577 | 62.7% | 73.7% |
| 70.0%–80.0% | 16 | 72.2% | 100.0% |
| 80.0%–90.0% | 0 | 0.0% | 0.0% |
| 90.0%–100.0% | 0 | 0.0% | 0.0% |