Table 5.

Feature selection stability in terms of Nogueira score [31] and Jaccard index (JAC) of the five feature selection methods. Whereas boosting and bagging used separately with RENT resulted in worse or similar stability, combing them into BB-RENT led to significantly improved stability. All pairwise comparisons for JAC between the five configurations showed significant differences.

Variable	Single EN	RENT	Boosted RENT	Bagged RENT	BB-RENT
Mean # features (n = 100)	30.6 ± 4.0	5.7 ± 1.5	24.9 ± 0.7	5.2 ± 1.6	6.6 ± 1.7
Nogueira score [31]	0.34	0.37	0.20	0.34	0.54
$\mathrm{JAC} (\mathrm{n}={}_2{}^{100}\mathrm{C}$ = 4950)	0.24 ± 0.06	0.25 ± 0.13	0.14 ± 0.05	0.23 ± 0.15	0.39 ± 0.14
t-test p-values [31]
RENT	0.047	-	-	-	-
Boosted RENT	<0.001	<0.001	-	-	-
Bagged RENT	0.962	0.162	<0.001	-	-
BB-RENT	<0.001	<0.001	<0.001	<0.001	-

Bold p-values indicate significant differences (p < 0.05) from the independent t-test. RENT = repeated elastic net technique, BB-RENT = bagged-boosted RENT, Mean # features = average number of features nominated across 100 runs, JAC = Jaccard index.