Passenger's train choice behavior is an important basis for railway operators to develop train operation plans. This paper uses a nonparametric machine learning method, namely support vector regression (SVR) to analyze and understand high-speed rail passengers’ train choice behavior including advance ticket purchase time, travel classes and train types, based on railway ticket booking data. Traditional choice models are often discrete choice models, which require a prior knowledge and assumptions on the relations between explanatory variables and choices. SVR has superb capabilities of extracting rules from data and does not need any prior knowledge. It can efficiently extract the rule of passengers’ choice behavior from the ticket booking data. In order to describe and predict the passenger's train choice behavior, this paper proposes a new metric γ incorporating the advance ticket purchase time. It can effectively reflect the acceptance of the train on the transport market by passengers. This new metric will be used as the response variable of the SVR model, and several train service attributes, such as day of week, departure time and travel time are identified as input variables. In addition, we also analyze the passengers’ train choice behavior in different periods. Finally, this paper conducts an empirical analysis of high-speed rail passengers along the Nanjing-Shanghai railway, considering multiple origin-destination pairs. Results demonstrate that SVR model can predict the train choice behavior with high accuracy. Keywords: High-speed, Choice behavior, Ticket data, Support vector regression

CICTP