Towards Application of One-Class Classification Methods to Medical Data

Table 2

AUC average and standard deviation, in brackets, values on UCI data sets. In the last column the distance used by the typicality method is indicated: c: correlation, E: Euclidean, E-st: Euclidean after standardization, and M: Mahalanobis.

Data sets	Target class	Gaussian	Mixture Gaussians	Naive Parzen	Parzen	Support vector DD	Typicality distance
Breast Wisconsin original	Benign	98.5 (0.1)	98.3 (0.2)	98.7 (0.1)	99.2 (0.1)	99.0 (0.1)	99.4 (0.2)—E
	Malignant	82.3 (0.2)	69.1 (3.2)	96.5 (0.4)	72.3 (0.5)	66.1 (0.8)	97.6 (0.5)—E
Breast Wisconsin prognostic	Returning	63.0 (1.4)	59.1 (1.6)	59.0 (1.9)	59.4 (1.9)	59.6 (1.4)	58.5 (5.4)—M
	Nonreturning	50.8 (0.8)	52.6 (1.6)	53.8 (2.2)	52.2 (1.7)	51.7 (1.7)	55.6 (2.9)—M
Colon	1	61.1 (3.8)	NaN	73.4 (3.1)	63.6 (22.4)	63.6 (22.4)	75.4 (6.3)—c
	2	70.4 (1.1)	NaN	70.0 (1.5)	36.4 (22.4)	36.4 (22.4)	78.3 (5.8)—c
E. coli	Periplasm	92.9 (0.3)	92.0 (0.4)	93.0 (0.8)	92.2 (0.4)	89.4 (0.8)	95.4 (1.3)—E
Hepatitis	Normal	82.1 (1.0)	78.3 (1.0)	80.1 (0.7)	79.0 (1.0)	78.7 (1.1)	80.8 (2.2)—M
Leukemia	1	92.1 (1.8)	NaN	90.2 (4.4)	NaN	58.9 (30.2)	91.2 (3.4)—c
	2	94.7 (2.7)	NaN	96.7 (0.4)	NaN	41.1 (30.2)	90.6 (3.9)—c
Liver disorders	Class 1	58.5 (0.4)	59.3 (0.7)	61.4 (0.7)	58.7 (0.4)	59.0 (0.9)	58.1 (2.5)—M
	Class 2	50.9 (0.5)	49.4 (0.6)	48.4 (0.8)	46.9 (0.8)	49.6 (1.0)	58.0 (3.7)—M
METAS	1	69.1 (1.5)	NaN	65.3 (0.8)	64.8 (21.5)	64.8 (21.5)	67.3 (2.3)—c
	2	36.4 (1.4)	NaN	40.7 (1.2)	35.2 (21.5)	35.2 (21.5)	64.5 (4.7)—c
SPECT heart	Class 0	93.4 (0.9)	95.1 (0.8)	90.7 (1.5)	95.7 (1.0)	89.7 (3.2)	86.1 (3.8)—M
	Class 1	28.4 (0.5)	27.9 (1.3)	26.0 (0.7)	44.5 (0.5)	57.1 (11.1)	69.8 (2.5)—M
Thyroid	Normal	84.3 (0.0)	84.7 (4.4)	96.1 (0.0)	90.6 (0.0)	56.0 (0.0)	98.1 (1.2)—c
	Hyperthyroid	70.3 (0.0)	68.1 (0.9)	75.1 (0.0)	70.6 (0.0)	45.7 (0.0)	65.9 (2.5)—c
	Subnormal	69.6 (0.0)	81.5 (1.0)	84.4 (0.0)	87.4 (0.0)	50.3 (0.0)	88.0 (2.8)—c