Research on the nonlinear relationship between street form and vitality in community-type subway station areas based on machine learning

Numerous studies have investigated the relationship between street spatial form and vitality. However, these studies often rely on small sample sizes, assume a linear response of crowd spatiotemporal behavior to urban form characteristics, and employ statistical methods with inherent limitations. In addition, the temporal variation in the influence of urban form on street vitality, particularly across different times of day, has not been sufficiently examined. To address these gaps, this study evaluates three dimensions of urban form, namely network form, interface form, and functional form, using 183 sets of street spatial data from representative community-based rail station areas in Beijing. It analyzes street vitality during representative measurement periods for three distinct hours during summer. Four machine learning models and two traditional linear regression models were constructed and compared. The results indicate that the Gradient Boosting Decision Tree model, which incorporates spatial information, achieves the highest accuracy. Feature importance analysis reveals that during the morning period, interface form variables, such as interface density and line alignment ratio, have a greater impact on crowd behavior, whereas during midday and nighttime, network and functional form variables, such as network density and functional diversity, are more influential. Partial Dependence Plot analysis further demonstrates that street spatial form characteristics influence crowd spatiotemporal behavior in complex, nonlinear patterns, with temporal variations in effect. These findings offer valuable guidance for urban planners seeking to optimize spatial layouts around rail station areas and predict patterns of crowd activity.