Investigation of factors that induce activities in green spaces of neighborhoods from the perspective of the behavioral change stage model using machine learning

Abstract

This study employed machine learning to generate hypotheses about the factors (and their combinations) that induce activities in green spaces in neighborhoods from the perspective of behavioral change. A cross-sectional study was conducted in March 2021 using an online survey in Osaka, Japan (n = 5108). We conducted (1) probabilistic latent semantic analysis (pLSA) to extract latent classes related to activities in green spaces and (2) classification tree analysis with the latent classes as the objective variables. As a result of clustering through pLSA, six latent classes were extracted for activities in green spaces. These classes indicated a gradation of behavioral change stages as follows: (1) difficult access, (2) precontemplation, (3) contemplation, (4) preparation, (5) action–maintenance, and (6) stability. Classification tree analysis indicated that different factors at different behavioral change stages promoted or inhibited activities in green spaces. “Difficult access” was more likely applicable when urban residents lived in neighborhoods where both the built environment and the social environment were undesirable. “Precontemplation” was more likely applicable where urban residents did not have positive perceptions of the effects of green space and social interactions with neighbors. “Contemplation” and “preparation” did not show any growth in the classification tree. “Action–maintenance” was more likely applicable when urban residents engaged in social activities in the neighborhoods. “Stability” was more likely applicable when urban residents had positive perceptions of the effects of green space and social interactions with neighbors.