Table 2.

Classification accuracies, sensitivity and specificity, sparsity and stability for all the network-based models compared, obtained with L2-norm SVM. The ^⁎ denotes a p-value < 0.05. p-values were obtained using permutation tests, as described in the main text.

Classification results L2-norm SVM
Features	Accuracy (%)	Accuracy p-value	Sensitivity (%)	Specificity (%)	Sparsity (%)	Stability (%)
Event-related fMRI dataset
Sparse inverse covariance	73.68	0.01⁎	73.68	73.68	–	37.81 ± 0.26
Full inverse covariance	44.73	> 0.05	47.37	42.11	–	9.90 ± 0.78
Correlation	68.42	0.02⁎	84.21	52.63	–	9.29 ± 1.56
Partial correlation	65.78	> 0.05	73.68	57.89	–	11.33 ± 0.16
Block-related fMRI dataset
Sparse inverse covariance	78.33	0.01⁎	80.00	76.67	–	25.81 ± 0.45
Full inverse covariance	40.00	> 0.05	20.00	60.00	–	12.81 ± 0.89
Correlation	60.00	> 0.05	86.67	33.33	–	8.61 ± 2.17
Partial correlation	58.33	> 0.05	53.33	63.33	–	10.93 ± 0.78

^⁎p-value < 0.05.