Skip to main content
. 2021 Dec 17;141:105153. doi: 10.1016/j.compbiomed.2021.105153

Table 4.

Hyperparameters of the pre-trained networks: Feature extraction hyperparameters were adopted from the optimal values in previous related work [20], while classifier hyperparameters were optimised on the pre-training data using cross-validation.

FEATURE EXTRACTION HYPERPARAMETERS
Hyperparameters Values
M MFCCs 39
F Frame length 210 = 1024
S
Segments
150
CLASSIFIER HYPERPARAMETERS
Hyperparameters
Classifier
Values
Convolutional filters CNN 256 & 128 & 64
Kernel size CNN 2
Dropout rate CNN, LSTM 0.2
Dense layer (for pre-training) CNN, LSTM, Resnet50 512 & 64 & 4
Dense layer (for fine-tuning) CNN, LSTM, Resnet50 32 & 2
LSTM units LSTM 512 & 256 & 128
Learning rate LSTM 10−3 = 0.001
Batch Size CNN, LSTM, Resnet50 27 = 128
Epochs CNN, LSTM, Resnet50 70