Calibrating a Bayesian transit assignment model using smart card data

Public transport planners can predict passenger loads and levels of service by applying the prior knowledge about the transit network and using transit assignment models. The individual travel history data available from automated fare collection (AFC) systems bring the opportunity of understanding the individual's travel behavior, which is necessary to develop a transit assignment model. By combining the prior knowledge about the transit network with the AFC data, a transit assignment model can be calibrated. This paper proposes a Bayesian hierarchical model to estimate attributes of travel time components and to calibrate a transit assignment model. In this model, route choices are represented by a multinomial logit model, and its coefficients are estimated via a Markov chain Monte Carlo method. The proposed model is specified in two ways, and in order to consider travel time variability, it is assumed that travel time on links follows a gamma distribution. In the first specification, route choice variables and parameters are the same for all transit modes of bus, train, and ferry. In the second specification, mode-specific route choice variables and parameters are defined. In order to assess the model fitness, the root-mean-square error (RMSE) between each posterior estimate and the actual observation is computed. The lowest %RMSE belongs to the third-model specification (at 15%), which indicates its high predictive power