Table 2.
Comparison of the different unsupervised outlier detection methods when applied to each subject separately.
| Statistical methods | HBOS | LOF | ABOD | OCSVN | LSCP |
|---|---|---|---|---|---|
| Orientation | 0.564 0.473 |
0.218 0.065 |
0.11 0.06 |
0.41 0.29 |
0.577 0.489 |
| Color | 0.5 0.4 |
0.241 0.091 |
0.1 −0.08 |
0.36 0.23 |
0.51 0.411 |
| Representation learning | AUTO | PCA | VAE | GAAL | MGAAL |
| Orientation | 0.53 0.44 |
0.527 0.426 |
0.477 0.368 |
0.429 0.311 |
0.428 0.309 |
| Color | 0.51 0.42 |
0.477 0.367 |
0.478 0.368 |
0.241 0.086 |
0.389 0.263 |
We calculated the mean f-score and Cohen's Kappa (first and second row in every cell) across all subject. HBOS, Histogram based outlier detection; LOF, Local outlier factor Method; ABOD, Angle-based outlier detector; OCSVM, One class support vector machine; LSCP, Locally selective combination of parallel outlier Ensembles; AUTO, Auto-encode based method; VAE, Variational auto-encoder based method; GAAL, Generative Adversarial Active Learning; MGAAL, Multi-object Generative Adversarial Active Learning.