Table 2.

Results from DL approach: different values of hyperparameters for CNNs (WaveNet, DenseNet, and ResNet). The hyperparameter patience indicates the number of epochs to wait before early stop if no improvement in the loss function is achieved. There is no optimal dilation rate value since it is only used for WaveNets, while the best model was a ResNet CNN.

Hyperparameter	Compared options
Kind of CNN	WaveNet, DenseNet, ResNet
Hidden layers number	4 8 12
Dilation rate (only for WaveNets)	1 2 4 8
Batch dimension	4 16 32 64 256
Kernel dimension	3 6 10 30 50 80
Activation function	ReLU ELU SELU
Optimizer	SGD ADAM NADAM
Start learning rate	${10}^{-1}$ ${10}^{-2}$ ${10}^{-3}$ ${\mathbf{10}}^{-\mathbf{4}}$ ${10}^{-5}$
Patience	5 8 10