Table 4.
Detailed performance metrics by eye colour and threshold. Results shown both for eye-colour defined through visual inspection and clustering algorithms.
| Colour | Threshold | Visual inspection eye-colour | Clustering-based eye-colour (k-means on non-normalized data) | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| ACC | Sensitivity | Specificity | PPV | NPV | ACC | Sensitivity | Specificity | PPV | NPV | ||
| Blue | 0 | 0.962 | 0.828 | 0.981 | 0.857 | 0.976 | 0.937 | 0.842 | 0.945 | 0.571 | 0.986 |
| Blue | 0.5 | 0.970 | 0.885 | 0.980 | 0.852 | 0.985 | 0.948 | 0.941 | 0.948 | 0.593 | 0.995 |
| Blue | 0.7 | 0.972 | 0.880 | 0.984 | 0.880 | 0.984 | 0.953 | 0.941 | 0.955 | 0.640 | 0.995 |
| Brown | 0 | 0.756 | 0.994 | 0.321 | 0.729 | 0.964 | 0.878 | 0.995 | 0.491 | 0.867 | 0.964 |
| Brown | 0.5 | 0.778 | 0.994 | 0.342 | 0.754 | 0.963 | 0.891 | 0.994 | 0.520 | 0.882 | 0.963 |
| Brown | 0.7 | 0.809 | 1.000 | 0.379 | 0.784 | 1.000 | 0.907 | 1.000 | 0.556 | 0.895 | 1.000 |
| Intermediate | 0 | 0.000 | 0.000 | 1.000 | 0.000 | 0.769 | 0.000 | 0.000 | 1.000 | 0.000 | 0.849 |
| Intermediate | 0.5 | 0.000 | 0.000 | 1.000 | 0.000 | 0.783 | 0.000 | 0.000 | 1.000 | 0.000 | 0.857 |
| Intermediate | 0.7 | 0.000 | 0.000 | 1.000 | 0.000 | 0.809 | 0.000 | 0.000 | 1.000 | 0.000 | 0.870 |