Latent mobility pattern estimation in the migration game

In human behavioral analysis, it is desired to accurately define basic quantities related to the decision-making process. In game theoretic models, essential such quantities are the utilities representing and quantifying behavioral preferences. Commonly, in order to estimate these quantities mobility patterns need to be extracted and the individual's decision behavior has to be traced. In this paper, we adopt the radiation model as a base to estimate the mobility pattern from population data only. And propose a method for extracting the latent mobility pattern using Bayesian inference, considering the spatial continuity of the mobility pattern and the time continuity of the population distribution change. We evaluate the accuracy of the estimated mobility pattern in the migration game, using the Japanese population census. We show that by incorporating the time-space continuity the model error of the radiation model is reduced. In addition, we calculate the utility from the estimated mobility pattern, and confirm that the proposed method is promising for extracting the game structure under limited information.