One-Class Classification by Ensembles of Random Planes (OCCERPs)

Table 3

Average AUCROC of various OCC algorithms against the OCCERP (500) algorithm on various datasets [28, 29] presented in Table 1. Bold numbers indicate the best performance.

Dataset	If	LOF	OCSVM	Autoencoder	OCCERP (500)
Pima	0.731	0.709	0.700	0.648	0.738
segment0	0.474	0.815	0.294	0.342	0.923
Winequality-red-4	0.584	0.651	0.615	0.609	0.655
Winequality-red-8_vs_6	0.667	0.592	0.647	0.681	0.716
Winequality-white-3_vs_7	0.849	0.866	0.853	0.851	0.928
yeast1	0.543	0.615	0.548	0.534	0.589
yeast3	0.673	0.807	0.725	0.728	0.788
yeast4	0.734	0.665	0.733	0.745	0.745
Aloi-unsupervised	0.539	0.748	0.549	0.549	0.556
Annthyroid-unsupervised	0.737	0.907	0.727	0.702	0.766
Breast-cancer-unsupervised	0.982	0.985	0.985	0.982	0.985
Letter-unsupervised	0.627	0.862	0.615	0.526	0.872
Satellite-unsupervised	0.949	0.977	0.937	0.895	0.977
Shuttle-unsupervised	0.995	0.999	0.996	0.993	0.999
Pen-global-unsupervised	0.947	0.957	0.972	0.869	0.998
Pen-local-unsupervised	0.778	0.985	0.589	0.440	0.966
Best performance	0	8	0	1	11