A comparison of feature extraction methods for prediction of neuropsychological scores from functional connectivity data of stroke patients

Brain Informatics

Table 2 Results with \(\alpha =0.001\) and \(\alpha =1\)

Cognitive domain	Method	\(\alpha =0.001\)				\(\alpha =1\)
Cognitive domain	Method	\({R}^{2}\)	MSE	BIC	NZ	\({R}^{2}\)	MSE	BIC	NZ
Language (\(n=95\))	PCA	0.52	0.48	404	45	0.49	0.51	365	35
	ICA	0.49	0.50	364	35	0.50	0.49	303	22
	DL	0.43	0.57	352	30	0.43	0.57	366	33
	NNMF	0.42	0.57	422	45	0.44	0.56	378	36
Spatial memory (\(n=78\))	PCA	0.23	0.76	396	45	0.23	0.76	295	22
	ICA	0.24	0.75	395	45	0.19	0.80	344	32
	DL	0.20	0.79	421	50	0.20	0.79	286	19
	NNMF	0.15	0.84	403	45	0.20	0.79	364	37
Verbal memory (\(n=78\))	PCA	0.27	0.72	501	70	0.34	0.65	319	30
	ICA	0.28	0.72	391	45	0.19	0.80	378	40
	DL	0.18	0.81	444	55	0.07	0.91	297	19
	NNMF	0.10	0.88	451	55	0.08	0.91	266	12

Performance of regularized regression with the \(\alpha\) parameter fixed at 0.001 (ridge) and 1 (LASSO) in the prediction of neuropsychological scores as a function of the feature extraction method. The number of non-zero features (NZ) is also reported. \({R}^{2}\): percentage of variance explained
MSE mean squared error, BIC Bayesian information criterion