Table 1.

Average number of features selected (α = 0.05) and number of corresponding groups and relevant groups on linear artificial datasets (500 variables, 50 groups, 5 relevant groups, and 100 samples) for each method.

		CER			CERr			eFDR			mProbes
		feat	gps	rel. gps	feat	gps	rel. gps	feat	gps	rel. gps	feat	gps	rel. gps
K = 1	RF	7.15	1.55	1.55	47.75	20.35	3.70	11.85	1.85	1.75	1.75	0.2	0.2
	avg	18.50	2.20	2.20	14.85	1.40	1.30	21.45	2.70	2.60	16.00	1.75	1.75
	∑	5	0.30	0.30	7.75	0.45	0.45	7.5	0.40	0.40	7.40	0.35	0.35
	max	14.90	1.60	1.60	28.35	3.10	2.55	17.45	1.75	1.65	11	1.10	1.10
$K=\sqrt{p}$	RF	7.05	1.45	1.45	61.10	23.80	3.80	11.20	2.05	1.75	11.05	1.65	1.65
	avg	19.80	2.75	2.75	25.90	2.65	2.15	22.55	3.15	3.05	20.45	2.70	2.70
	∑	16.35	1.40	1.40	23.55	2.20	2.15	17.35	1.55	1.55	22.75	2.00	2.00
	max	12.50	1.65	1.65	35.90	4.00	2.90	14.50	1.80	1.75	12.95	1.75	1.75

RF means Random Forests without any aggregation function. Bold text and underlined text are for best number of relevant groups over all aggregation functions and over all selection methods respectively.