In recent years, it is one of the public interests to grasp the number of people who have difficulty in returning home after a devastating earthquake. In this paper, we construct a hierarchical neural network model, which describes people's decision making on returning home. The proposed model is calibrated using data taken from questionnaire surveys, and applied to people staying in Tokyo metropolitan area. We attempt to estimate the number of people who cannot return home if traffic networks are blocked by a devastating earthquake, and also demonstrate the spatio-temporal distribution of people who have difficulty in returning home.
