A Robust Supervised Variable Selection for Noisy High-Dimensional Data

Table 2

Leave-one-out cross validation performance evaluated by classification accuracy for the data of Sections 4.1, 4.2, and 4.3. MRRMRR uses as the relevance measure and as the redundancy measure.

Dimensionality reduction	Classification method	Classification accuracy
		Section 4.1	Section 4.2	Section 4.3
—	SVM	1.00	1.00	0.93
—	Classification tree	0.94	0.97	0.55
—	LDA	Infeasible	Infeasible	Infeasible
—	PAM	0.85	0.98	0.75
—	SCRDA	1.00	1.00	0.79
Number of principal components		10	20	4
PCA	SVM	0.75	1.00	0.90
PCA	Clas. tree	0.72	0.97	0.59
PCA	LDA	0.57	0.90	0.79
PCA	PAM	0.64	0.81	0.77
PCA	SCRDA	0.71	0.92	0.79
Number of variables for MRRMRR		10	20	4
MRRMRR	SVM	1.00	1.00	0.93
MRRMRR	Clas. tree	0.76	0.97	0.55
MRRMRR	LDA	0.95	1.00	0.79
MRRMRR	PAM	0.82	0.97	0.75
MRRMRR	SCRDA	1.00	1.00	0.79