Table 3.

ADNI Dataset: comparisons of 5 repetitions of a nested 10-fold cross-validation balanced accuracy using the clinical information selected by a FDR procedure. The results are divided in 5 families: Baseline, Feature Selection (FS), Random Forests-based family (RF), standard Multiple Kernel Learning (MKL), Feature Weighting by using MKL (FW) and our method Feature Weighting and Selection (FWS). R corresponds to the number of kernels used.

	Algorithm	Kernels	R	Bal. Acc. %
Baseline	SVM	$\mathbf{I}$ + $\mathbf{C}$	1	$84.10\pm 7.92$
FS	SVM RFE	$\mathcal{V}$ & $\mathcal{C}$	$-$	$86.53\pm 5.99$
	SVM t-test	$\mathcal{V}$ & $\mathcal{C}$	$-$	$86.01\pm 5.17$
RF	Gray	$\mathbf{I}$ & $\mathbf{C}$	$-$	$85.99\pm 10.73$
	Pustina	$\mathbf{I}$ & $\mathbf{C}$	$-$	$84.34\pm 11.14$
MKL	SimpleMKL	$\mathbf{I}$ & $\mathcal{C}$	26	$84.29\pm 11.78$
	EasyMKL	$\mathbf{I}$ & $\mathcal{C}$	26	$84.47\pm 7.28$
FW	SimpleMKL	$\mathcal{V}$ & $\mathcal{C}$	168155	Out of memory
	EasyMKL	$\mathcal{V}$ & $\mathcal{C}$	168155	$87.97\pm 6.59$
FWS	EasyMKLFS	$\mathcal{V}$ & $\mathcal{C}$	168155	$92.38\pm 7.27$