Table 4.

Comparison of discriminative power and compactness for various regularizing methods.

Met hods	Discriminability				Interpretability		Mean squared error
	AD vs MCI	MCI vs.CN	AD vs. CN	AD+MCI vs CN	Sparsity	Overlap	Brain M	NPT tasks X
C-NMF	0.8108 (0.029)	0.7557 (0.0303)	0.9055 (0.0247)	0.8045 (0.0259)	0.7459 (0.0059)	0.2441 (0.008)	0.0028 (0.0001)	0.0 (0.0)
C-NMF + SV M	0.8059 (0.0335)	0.7404 (0.0601)	0.9056 (0.0224)	0.7938 (0.0443)	0.7507 (0.0089)	0.2395 (0.0099)	0.0029 (0.0001)	0.0 (0.0)
C-NMF + $l_1$norm	0.8035 (0.0338)	0.7981 (0.0493)	0.9001 (0.0283)	0.8287 (0.0406)	0.9189 (0.0066)	0.0611 (0.0065)	0.0055 (0.0002)	0.0003 (0.0)
C-NMF + $l_1$norm +SVM	0.7951 (0.0453)	0.8008 (0.0693)	0.9142 (0.0306)	0.8368 (0.0635)	0.9191 (0.0058)	0.0631 (0.0075)	0.0056 (0.0002)	0.0003 (0.0001)

We computed the average and standard deviation after 10 random resamplings of the train/test cohort.