2.3.2 The hyperparameters of the 1D-CNN model

The hyperparameters are generally classified into two types, including network structure-related and network training-related (Kolar et al., 2021). Among them, the main hyperparameters related to network structure include the number of convolution kernels—the number of filters, convolutional kernel size—the filter size, number of hidden layers—a layer of neurons between the input and output layers, dropout—random deactivation of a certain percentage of neurons, and activation function—whether a neuron should be activated or not. The other hyperparameters include loss function—a measure of how far the predictions deviate from the true value, batch size—the selection of a sample set to update the weights, the number of iterations—the number of times the entire process is repeated, and learning rate—an adjustment parameter in optimization algorithms.

Hyperparameter selection has a significant impact on the performance of the detection model, which takes more time and requires an enriching experience with manual hyperparameter tuning. However, it is difficult to find the optimal set of hyperparameters through manual experience alone; hyperparameter selection can be quickly searched using intelligent optimization algorithms.