Variational Bayesian Inference for Time Series Route Choice Model for Evaluating Flexible Measures for Time and Location

Recently, flexible measures for time and location attract attention because people’s places of activity have become more diverse. This study addresses the temporal changes in pedestrians’ travel preferences in route choice models for the evaluation of the measures. We extend the conventional model to the time series direction by setting different parameters for each time period and develop an estimation method utilizing variational inference to reduce the computational load of model estimation. We confirm that our model has higher accuracy than the conventional model and that our estimation method can reduce the estimation time to 1/31 of that of the conventional estimation method. Furthermore, we present an example of the application of our model through the evaluation of the measures.