GAT-LI: a graph attention network based learning and interpreting method for functional brain network classification

Table 4 Classification performance on networks with different sparsity (mean \(\pm\) std)

Threshold	Number of edges	Sparisty	Accuracy	Sensitivity	Specificity	F1	AUC
0.1	10,056	0.1689	0.6686 ± 0.0344	0.7287 ± 0.0358	0.6113 ± 0.0790	0.6826 ± 0.0236	0.7396 ± 0.0317
0.2	7976	0.3408	0.6512 ± 0.0604	0.7327 ± 0.0360	0.5736 ± 0.1110	0.6738 ± 0.0433	0.7165 ± 0.0603
0.3	5877	0.5142	0.6377 ± 0.0684	0.7149 ± 0.0449	0.5642 ± 0.1255	0.6600 ± 0.0467	0.7005 ± 0.0648
0.4	3933	0.6750	0.6232 ± 0.0693	0.7129 ± 0.0429	0.5377 ± 0.1303	0.6506 ± 0.0460	0.6878 ± 0.0681
0.5	2330	0.8074	0.6145 ± 0.0516	0.6594 ± 0.0409	0.5717 ± 0.0992	0.6263 ± 0.0370	0.6831 ± 0.0542
Dense network	12,100	0	0.6802 ± 0.0269	0.7406 ± 0.0408	0.6226 ± 0.0534	0.6931 ± 0.0248	0.7358 ± 0.0373

ISSN: 1471-2105