Freight Mode Choice Models Using Artificial Neural Networks

This paper presents a new approach to behavioral choice modeling using artificial neural networks (ANNs). A feed-forward network trained by a back-propagation learning algorithm is used in this study. As a modeling technique, ANNs are highly adaptive and very efficient in dealing with problems involving complex interrelationships among many variables. The application of ANNs in the development of mode choice models is tested on the U.S. freight transport market using information on individual shippers and individual shipments. Shipments are disaggregated at the 5-digit Standard Transportation Commodity Code (STCC) level, representing the most detailed information publicly available. Results obtained from this exercise are compared with similar results obtained from conventional logit and probit disaggregate mode choice models. ANNs produced slightly better results compared with both logit and probit models. A method for analyzing ANN results based on examining variable link weights is described. The method allows for increasing the efficiency of ANNs by selecting only those input variables which significantly contribute to the network output. ANN mode choice models are expected to behave equally well in the passenger transport market, both in the urban and intercity travel contexts.