A Customized Deep Neural Network Approach to Investigate Travel Mode Choice with Interpretable Utility Information

Algorithm 1

Training the model.

	Input The feature vectors, .
	Observations of the individual choice, .
	Output The model with learnt parameters.
(1)	Procedure DNN model Train.
(2)	Initialize the parameter matrix: for embedding.
(3)	Embedding categorical values: .
(4)	Initialize a null set: .
(5)	for all available individual sample do.
(6)	.
(7)	.
(8)	A training sample is placed in .
(9)	end for
(10)	Initialize all the weight and intercept parameters.
(11)	Initialize , for BN.
(12)	repeat.
(13)	Randomly extract a batch of samples from .
(14)	Update the parameters by minimizing the equation (7) by the mini-batch gradient descent algorithm within .
(15)	until convergence criterion is met.
(16)	end procedure.