Environmental determinants of dynamic jogging patterns: Insights from trajectory big data analysis and interpretable machine learning

Abstract

The dynamic patterns of leisure jogging profile the jogger-environment interactions. However, the dynamic patterns and their nonlinear associations with environmental factors are poorly explored. Therefore, we develop a framework to uncover the dynamic jogging patterns and interpret their nonlinear and interactive associations with environments. Initially, the bivariate time series clustering method discerns daily and weekly patterns from the integrated jogging flow and duration. Then, interpretable machine learning methods including CatBoost, SHAP, and ALE plots elucidate the nonlinear and interactive relationships. An empirical analysis of Beijing, China was conducted using multisource data. Our finding highlights that (1) five distinct daily and weekly jogging patterns were investigated for area zoning. These patterns show notable spatial-temporal disparities in jogging flow and duration. (2) Built environment (BE) and visual environment are crucial in shaping jogging, with accessibility and facilities being significant contributors. (3) Environmental variables show significant nonlinear and threshold effects on leisure jogging, which vary across jogging patterns and urban areas. (4) Interaction effects among environmental factors were investigated. BE factors like sports amenity exert more significant interactions. Importantly, incorporating geographic locations enhances model performance as it captures spatial effects. These findings can help planners design refined intervention strategies for leisure activities.