Skip to main content

Table 3 Overall performance of schizophrenic speech detection using Sch-net and its components (backbone, skip connection (SC), and CBAM)

From: Sch-net: a deep learning architecture for automatic detection of schizophrenia

Evaluated indicators 95% CI
Backbone Backbone + SC Backbone + CBAM Sch-net (ours)
Accuracy 0.9323 0.9494 0.9563 0.9768
(0.9295,0.9351) (0.9460,0.9528) (0.9534,0.9591) (0.9739,0.9797)
Precision 0.9480 0.9634 0.9513 0.9639
(0.9445,0.9515) (0.9564,0.9704) (0.9458,0.9568) (0.9585,0.9693)
Recall 0.9149 0.9348 0.9622 0.9908
(0.9100,0.9197) (0.9326,0.9370) (0.9556,0.9688) (0.9898,0.9918)
F1-score 0.9311 0.9487 0.9565 0.9771
(0.9280,0.9341) (0.9456,0.9519) (0.9536,0.9594) (0.9743,0.9799)
Sensitivity 0.9176 0.9619 0.9902 0.9914
(0.9131,0.9221) (0.9581,0.9657) (0.9847,0.9956) (0.9863,0.9964)
Specificity 0.9488 0.9601 0.9494 0.9738
(0.9415,0.9561) (0.9513,0.9689) (0.9437,0.9551) (0.9656,0.9820)
AUC 0.9593 0.9892 0.9902 0.9978
(0.9577,0.9609) (0.9859,0.9924) (0.9880,0.9924) (0.9965,0.9990)