Skip to main content

Table 4 Performance of feature engineering and classifiers on schizophrenic speech detection

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

Classifier   Feature
Time-domain feature FFT-based spectral feature Auditory-based spectral feature Spectral envelope feature
STE Pitch Fluency feature LTAS Spectrogram MFCC GTCC LP SWLP XLP
RF Accuracy 0.7686 0.5935 0.8213 0.6464 0.8972 0.8043 0.8791 0.9245 0.9377 0.9423
Precision 0.6251 0.5847 0.8281 0.6052 0.8946 0.7818 0.8487 0.9055 0.9319 0.9282
Recall 0.7126 0.5754 0.8103 0.5545 0.9103 0.8577 0.9289 0.9549 0.9466 0.9644
F1-score 0.6306 0.6322 0.8133 0.5513 0.8972 0.8144 0.8856 0.9280 0.9391 0.9453
KNN Accuracy 0.7723 0.5385 0.7390 0.7504 0.8974 0.8626 0.8753 0.9204 0.9287 0.9375
Precision 0.6410 0.5308 0.7050 0.6489 0.8566 0.8977 0.9043 0.8939 0.9117 0.9196
Recall 0.6063 0.5597 0.7985 0.6152 0.9636 0.8312 0.8494 0.9640 0.9549 0.9636
F1-score 0.6123 0.5382 0.7418 0.6211 0.9046 0.8591 0.8700 0.9257 0.9315 0.9398
SVM Accuracy 0.7905 0.5172 0.7746 0.7358 0.9024 0.8625 0.8929 0.9164 0.9291 0.9334
Precision 0.6447 0.5087 0.7657 0.6556 0.8741 0.8555 0.8762 0.8980 0.9183 0.9126
Recall 0.5999 0.4767 0.7875 0.5435 0.9549 0.8929 0.9198 0.9470 0.9466 0.9636
F1-score 0.6155 0.4644 0.7627 0.5813 0.9091 0.8689 0.8960 0.9206 0.9317 0.9356
LDA Accuracy 0.7858 0.5087 0.7452 0.7314 0.8385 0.8887 0.9198 0.9026 0.9069 0.9109
Precision 0.6447 0.4625 0.7083 0.6622 0.8053 0.9394 0.9479 0.8963 0.9053 0.8868
Recall 0.5898 0.5391 0.7522 0.5380 0.9095 0.8474 0.8933 0.9198 0.9111 0.9462
F1-score 0.6093 0.4710 0.7104 0.5821 0.8498 0.8807 0.9161 0.9060 0.9079 0.9146