Comparison of linear discriminant analysis methods for the classification of cancer based on gene expression data

Table 3 Average test error of LDA and its modification methods (10 cycles of 10-fold cross validation)

Dataset	Gene selection methods	Performance
		LDA	PAM	SDDA	SLDA	SCRDA
2-class Lung cancer data(n = 181, p = 12533, K = 2)	PAM	0.30	0.26	0.15	0.16	0.42
	SDDA	0.17	0.11	0.1	0.11	0.1
	SLDA	0.47	0.3	0.3	0.3	0.32
	SCRDA	0.73	0.20	0.19	0.17	0.19
Colon data(n = 62, p = 2000, K = 2)	PAM	1.30	0.82	0.8	0.86	0.86
	SDDA	2.25	2.09	1.33	1.29	1.25
	SLDA	1.12	0.74	0.75	0.77	0.80
	SCRDA	1.19	0.77	0.77	0.75	0.78
Prostate data(n = 102, p = 6033, K = 2)	PAM	2.87	0.89	0.82	0.81	1.00
	SDDA	2.53	0.71	0.72	0.68	0.74
	SLDA	1.75	0.7	0.64	0.64	0.70
	SCRDA	2.15	0.57	0.59	0.57	0.61
Multi-class lung cancer data(n = 66, p = 3171, K = 6)	PAM	2.13	1.16	1.21	1.28	1.19
	SDDA	1.62	1.32	1.32	1.31	1.30
	SLDA	1.62	1.31	1.32	1.26	1.34
	SCRDA	1.63	1.43	1.45	1.58	1.35
SRBCT data(n = 83, p = 2308, K = 4)	PAM	0.17	0.01	0.01	0.03	0.01
	SDDA	2.45	0.03	0.02	0	0.03
	SLDA	2.87	0	0	0	0
	SCRDA	2.32	0.03	0.03	0.02	0.03
Brain data(n = 38, p = 5597, K = 4)	PAM	1.14	0.57	0.57	0.58	0.61
	SDDA	1.09	0.61	0.62	0.63	0.55
	SLDA	0.89	0.60	0.60	0.57	0.58
	SCRDA	0.84	0.56	0.54	0.54	0.57

ISSN: 1756-9966