Table 2.

Classification performance.

Classification method
$i$	lasso	ridge	wknn	nb	svm	gpls	nnet	cart	c5.0	rf	gbt
1	0.867 (185)	0.864 (185)	0.853 (185)	0.852 (185)	0.851 (185)	0.838 (185)	0.822 (185)	0.795 (185)	0.795 (185)	0.864 (185)	0.862 (185)
2	0.856 (89)	0.847 (86)	0.845 (98)	0.849 (70)	0.530 (5)	0.836 (80)	0.807 (117)	0.799 (106)	0.803 (103)	0.864 (109)	0.855 (89)
3	0.857 (50)	0.854 (51)	0.845 (65)	0.829 (38)	0.537 (4)	0.836 (47)	0.809 (87)	0.794 (66)	0.803 (62)	0.866 (99)	0.859 (52)
4	0.856 (24)	0.853 (31)	0.837 (40)	0.832 (26)	0.542 (3)	0.838 (24)	0.801 (59)	0.799 (45)	0.790 (39)	0.865 (85)	0.858 (38)
5	0.853 (17)	0.853 (21)	0.842 (28)	0.838 (15)	0.562 (2)	0.838 (16)	0.793 (45)	0.811 (34)	0.806 (24)	0.865 (77)	0.855 (24)
6	0.854 (10)	0.851 (15)	0.847 (16)	0.841 (13)	—	0.837 (9)	0.810 (25)	0.817 (28)	0.803 (23)	0.863 (75)	0.856 (16)
7	0.850 (6)	0.854 (11)	0.833 (9)	—	—	0.838 (6)	0.812 (21)	0.822 (24)	0.804 (16)	0.864 (69)	0.854 (14)
8	—	0.854 (9)	0.829 (7)	—	—	—	0.852 (12)	0.822 (23)	0.802 (13)	0.865 (64)	0.853 (11)
9	—	0.854 (8)	0.830 (6)	—	—	—	0.857 (8)	0.822 (22)	0.802 (12)	0.865 (59)	—
10	—	0.853 (7)	—	—	—	—	0.842 (4)	—	0.809 (10)	0.866 (57)	—
11	—	—	—	—	—	—	—	—	—	0.865 (56)	—
12	—	—	—	—	—	—	—	—	—	0.864 (51)	—
13	—	—	—	—	—	—	—	—	—	0.864 (50)	—
14	—	—	—	—	—	—	—	—	—	0.863 (47)	—

Mean cross-validation AUC for each classifier with best parameter configuration and for each iteration ($i$). The number of features are given in parenthesis. The best run per classifier is highlighted in boldface. All methods induce at least one model with AUC of 0.809 or higher. Empty cells indicate that the feature selection wrapper had already been terminated after a previous iteration.