Identification of Driver Distraction Based on SHRP2 Naturalistic Driving Study

Table 3

Related parameters of classification models.

LSTM-NN
Type	Neurons numbers	Iterations	Batch size	Output layer activation function	Optimizer	Learning rate	Loss function
2	64	100	32	Sigmoid	Adam	10^–3	binary_crossentropy
4	64	200	32	Softmax		10^–2	categorical_crossentropy
SVM
Type	Main parameters		Adjusting parameters method			Optimum parameters
2	C, gamma, and kernel		5-fold cross-validation			C = 10^3, gamma = 10^−3, and kernel = rbf
4											C = 10^2, gamma = 10^−2, and kernel = rbf
AdaBoost
Type	Main parameters		Adjusting parameters method			Optimum parameters
2	Learning rate and n_estimation		5-fold cross-validation			Learning rate = 0.1 and n_estimation = 50
4											Learning rate = 0.15 and n_estimation = 100