A Triplet Multimodel Transfer Learning Network for Speech Disorder Screening of Parkinson’s Disease

Table 5

Performance of deep models on the MDVR-KCL dataset.

CNN-based methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)	Temporal network-based methods	Precision (%)	Recall (%)	F1 score (%)	Acc (%)
CNN	89.38	57.50	69.99	80.01	HMM	61.74	50.48	55.54	67.73
DNN [41]	80.76	51.26	62.71	76.06	LSTM	84.28	59.39	69.68	79.37
ResCNN [43]	63.40	46.09	53.38	68.13	LSTM (Attn)	88.42	51.08	64.76	77.79
ThinResNet [45]	86.96	59.60	70.73	79.96	BiGRU (Attn) [47]	92.45	64.49	75.98	84.04
DenseCNN [42]	85.71	86.84	87.16	86.47	BiLSTM(Attn) [46]	92.77	67.27	77.99	85.07
ResNet50 [44]	65.00	78.00	70.91	59.75	TmmNet	100.00	75.26	85.88	90.23

The bold values in Table 5 represents the highest results compared with the deep learning models.