Table 7.
Effect of SVEB-aware calibration on classification performance.
| Decision rule | Accuracy | Macro-F1 | SVEB precision | SVEB recall | SVEB F1 |
|---|---|---|---|---|---|
| Argmax baseline | 94.14% | 72.52% | 48.80% | 34.42% | 40.37% |
| SVEB-calibrated | 94.15% | 74.53% | 48.50% | 43.95% | 46.11% |
Calibration parameters were selected on the validation set and evaluated on the held-out test set.
Bold values indicate the metrics that improved after applying the selected SVEB-aware calibration compared with the argmax baseline.