A Robust Supervised Variable Selection for Noisy High-Dimensional Data

Table 1

Leave-one-out cross validation performance of various classification methods for the data of Section 4.1. MRMR is used in version (1) or (20) to find 10 variables, while the optimal over all is used. Sensitivity (SE) and specificity (SP) are given for selected fixed values of .

Dimensionality reduction			Classif. method	Classif. accuracy
MRMR variable selection
Measure of		MRMR criterion				Parameter
relev.	redund.					0	0.1	0.2	0.3	0.5	0.7	0.9
Mutual info.	Mutual info.	(1)	LDA	0.92	SE	0.75	0.83	0.92	0.88	0.96	0.96	0.96
					SP	0.67	0.92	0.88	0.92	0.96	0.92	0.92
		(1)	LDA	1.00	SE	0.92	0.92	0.83	0.88	0.96	0.96	0.96
					SP	0.88	0.96	0.96	0.96	0.96	1.00	1.00
		(1)	LDA	0.96	SE	0.83	0.83	0.96	0.83	0.92	0.96	0.96
					SP	0.88	0.88	0.83	0.96	1.00	0.96	1.00
	K-S	(1)	LDA	0.82	SE	0.92	0.92	0.92	0.92	0.92	0.88	0.88
					SP	0.88	0.88	0.88	0.88	0.88	0.96	0.96
	Sign test	(1)	LDA	0.82	SE	0.92	0.92	0.92	0.92	0.92	0.88	0.88
					SP	0.88	0.88	0.88	0.88	0.88	0.96	0.96
		(20)	LDA	1.00	SE	0.92	0.92	0.88	0.88	0.92	0.96	1.00
					SP	0.88	0.96	0.96	0.96	0.96	0.96	1.00
		(20)	LDA	1.00	SE	0.92	0.92	0.96	0.96	0.96	0.96	1.00
					SP	0.88	0.88	0.88	0.88	0.92	0.96	1.00
		(20)	LDA	1.00	SE	0.92	0.92	0.96	0.96	0.96	0.96	1.00
					SP	0.88	0.88	0.92	0.92	0.92	0.96	1.00
		(20)	LDA	1.00	SE	0.92	0.92	0.96	0.96	0.96	0.96	1.00
					SP	0.88	0.88	0.92	0.92	0.96	0.96	1.00