Table 3.
Performance on the Kahraman benchmark
| Method | AUC | CE |
|---|---|---|
| sup-CK | 0.858 ± 0.14 | 0.36 |
| sup-CKL | 0.861 ± 0.13 | 0.27 |
| sup-CK-Vol | 0.889 ± 0.14 | 0.34 |
| sup-CKL-Vol | 0.895 ± 0.12 | 0.26 |
| Vol | 0.875 ± 0.14 | 0.39 |
| Princ-Axis | 0.853 ± 0.13 | 0.35 |
| sup-PI | 0.815 ± 0.13 | 0.42 |
| SHD | 0.770 | 0.39 |
| MultiBind | 0.715 ± 0.17 | 0.42 |
| Sequence | 0.55 ± 0.08 | 0.8 |
Performance for each algorithm is evaluated by its mean AUC score and by its classification error (CE), averaged over all pockets. (AUC score for SHD are taken directly from [2], CE scores are estimated from data provided by authors).