Probabilistic inference on the factors promoting park use by machine learning using panel data from 2014 to 2020 in Tokyo, Japan

The health risks associated with confinement and physical inactivity during the COVID-19 pandemic continue to be a concern in urban areas. Access to urban green spaces is expected to contribute to increased physical activity. However, park use is the outcome of complex interactions among environmental, social, and individual factors. This study aimed to infer probabilistically factors promoting park use through hypothesis-generating research using two-point panel data analysis and machine learning before and after the COVID-19 epidemic in 2014 and 2020. We performed probabilistic inference using a structural model for probabilistic latent semantic analysis (pLSA) that combined pLSA and Bayesian network. The result indicated a slight but significant decrease in the frequency of park use between 2014 and 2020. However, the impact of COVID-19 on living condition and lifestyle was not the largest factor that influenced the frequency of park use in 2020 after COVID-19. Instead, we found that the presence of family members and social interaction with neighbors influenced the largest determinants of park use. The best social prescriptions for realistically and rapidly eliminating health disparities during the COVID-19 epidemic would be the implementation of interventions that simultaneously induce social capital improvement and park use. •Probabilistic inference on park use was estimated using machine learning.•A model of probabilistic latent semantic analysis and Bayesian network was used.•COVID-19 was not the largest factor of changes in the frequency of park use.•Conditions of social interactions with neighbors influenced park use.