Table 5.

Comparison of the proposed method with fully supervised state-of-the-art method.

Method	AUROC		AUPRC
	FNUSA	MAYO	FNUSA	MAYO
Fully supervised Nejedly et al.36 (tens of thousands training samples)	0.92	0.97	0.80	0.93
Proposed method with 1000 KDE training samples per class	0.895	0.896	0.794	0.749
Proposed method with 100 KDE training samples per class	0.862	0.879	0.740	0.714