Deep dive into model performance and calibration
Overall Accuracy
38.8%
31 / 80 correct
Total Predictions
80
Completed fights reviewed
High Confidence Accuracy
45.5%
Top tier predictions
High
45.5%
11 fights
Low
37.5%
8 fights
Medium
50.0%
22 fights
Uncertain
30.8%
39 fights
Accuracy shows how often the predicted winner won. Confidence tiers group predictions by calibrated probability. Higher confidence tiers should have higher accuracy.
| Tier | Total Fights | Correct | Accuracy | Avg Probability |
|---|---|---|---|---|
| High | 11 | 5 | 45.5% | 74.2% |
| Low | 8 | 3 | 37.5% | 63.8% |
| Medium | 22 | 11 | 50.0% | 64.8% |
| Uncertain | 39 | 12 | 30.8% | 54.5% |
| Version | Predictions | Correct | Accuracy |
|---|---|---|---|
| XGBoost (calibrated) | 52 | 31 | 59.6% |
Predicted probability vs actual win rate. Points on the diagonal indicate perfect calibration.
Calibration explained: The chart shows predicted probability (x-axis) vs actual win rate (y-axis). Points on the diagonal (45°) indicate perfect calibration - the predicted probabilities match observed outcomes. Points above the line indicate the model is overconfident. Points below indicate underconfidence.