Table 2.

Blind accuracies for various combinations of filtering and modeling methods

Filtering methods	Wrapper methods
	BFCS	FNN	SVM	MRA	kNN	WV
PART	81.1 ± 14.1 (8)	64.4 ± 15.6 (3)	73.3 ± 5.4 (2)	60.0 ± 16.6 (11)	67.8 ± 16.8 (10)	56.7 ± 13.6(14)
NSC	68.9 ± 6.7 (5)	60.0 ± 12.4 (3)	62.2 ± 13.3 (3)	65.6 ± 9.2 (9)	68.9 ± 13.0 (3)	66.7 ± 9.9 (21)
S2N	68.9 ± 6.7 (15)	56.7 ± 16.1 (3)	61.1 ± 15.9 (3)	61.1 ± 14.3 (4)	63.3 ± 17.2 (4)	58.9 ± 12.2 (18)
SAM	71.1 ± 7.4 (12)	64.4 ± 12.0 (3)	67.8 ± 13.6 (3)	63.3 ± 12.2 (10)	74.4 ± 8.7 (7)	63.3 ± 11.2 (9)
Student's t-test	71.1 ± 5.4 (15)	53.3 ± 12.0 (4)	60.0 ± 10.2 (13)	58.9 ± 13.2 (5)	68.9 ± 8.3 (4)	60.0 ± 19.4 (26)
U-test	66.7 ± 9.9 (9)	56.7 ± 16.1 (3)	64.4 ± 13.0 (7)	54.4 ± 10.2 (7)	67.8 ± 11.6 (14)	62.2 ± 12.4 (1)
Welch's t-test	65.6 ± 10.5 (15)	55.6 ± 14.9 (3)	58.9 ± 8.7 (13)	53.3 ± 12.2	67.8 ± 10.5	65.6 ± 13.6 (12)
No filtering	68.9 ± 9.7 (10)	58.9 ± 10.0 (3)	66.7 ± 15.7 (2)	61.1 ± 13.4 (3)	55.6 ± 15.7 (3)	57.8 ± 17.1 (26)

Parenthesized values indicate the numbers of probes used in each model.