Knowledge-Based Neural Networks and its Application in Discrete Choice Analysis

Travel mode choice forecast has received wide attention in travel behavior analysis. Mode choice is a pattern recognition problem, where different human behavior patterns determine the choices among alternative travel modes. Based on the functional similarity between artificial neural networks (ANN) and decision tree, the method of knowledge-based neural networks (KBNN) combines the rule induction of decision tree (DT) and the accurate approximation of ANN. One appeal of KBNN is the use of pattern association and error correction to represent a problem. This contrasts considerably with the random utility maximization framework in discrete choice modeling. So a network built by this method and a nested logit (NL) model are specified, estimated and comparatively evaluated. The prediction results show that KBNN model demonstrates the highest performance. The analysis of actual investigation data shows that the proposed KBNN model has fast convergence and high precision, which is of great importance for travel mode choice prediction.