Travel Mode Choice Modeling with Support Vector Machines

Abstract

This study investigates the applications of nontraditional models for travel mode choice modeling, which traditionally has relied on disaggregate discrete choice models such as multinomial logit models. A new artificial intelligence model, a support vector machine, is applied for the first time to travel mode choice modeling. This support vector machine model is tested and compared with a multinomial logit model and a multilayer feedforward neural network model based on data collected in the San Francisco Bay Area in California. Two scenarios with different training data sizes are tested. For both scenarios, the support vector machine model outperforms the multinomial logit model in terms of fitting and testing results. Although the multilayer feedforward neural network model performs best for fitting, it underperforms the other two models for testing. It is recommended that the support vector machine model be used as an alternative procedure for travel mode choice modeling because of its promising performance and easy implementation.