ESTIMATION OF STAYING BEHAVIOR ALONG LRT LINES USING INVERSE REINFORCEMENT LEARNING

Abstract

Amid the problem of declining public transportation in rural areas, the “Compact + Network” initiative is being promoted. Toyama City in Toyama Prefecture, which introduced the LRT, has seen an increase in the number of public transportation users and the rate of outings by residents along its rail line. On the other hand, a quantitative evaluation from a more microscopic viewpoint is required for the impact of the LRT on the liveliness along its rail line. In this study, inverse reinforcement learning, a type of machine learning, was used to generate action trajectories from smartphone location data, and estimated the behavior of people staying along LRT lines. The model was able to evaluate the characteristics of location data with high accuracy, and the model showed that there are more staying behaviors along LRT lines than along non-lines, and that the staying behaviors decay over time. The analysis suggests the importance of environmental improvement along LRT lines in urban development utilizing LRT.