Travel behavior model for traffic micro-simulation system: a machine learning approach

Machine learning approach describes data using statistical theory which is not constrained by economic explanation, and therefore, can be better to describe behavioral data. However, a conventional machine learning model (e.g. artificial neural network model) is not suitable for processing the stated preference (SP) / revealed preference (RP) survey data which include multiple choice set patterns. This paper presents a new machine learning model for predicting travel behavior. The proposed model handles the SP/RP survey data and also can provide choice probability information. A case study was carried out by using the proposed model and logit model. The result indicates that the proposed model can fit the data better. (a) For the covering entry of this conference, please see ITRD abstract no. E217226.