A Comparison of Cross-Nested Logit Model and BP Neural Network to Estimate Residential Location and Commute Mode Choice in Beijing

The objective of this paper is to compare the merits of back propagation neural network (BPNN) with those of cross-nested logit (CNL) model to estimate the simultaneously joint choice of residential location and commute mode choice during the process of employment surburbanization. Back propagation neural network and discrete choice model specified as cross-nested logit have been respectively employed to investigate the joint choice for different types of employment destination scenarios, that is, under center (CBD), urban and suburban workplace patterns in Beijing. The predictive capability of these two models has been compared in terms of models accuracy. Results demonstrate that on the whole the BPNN have a higher accuracy for this joint choice and is more suitable for prediction.