Interpretable Machine Learning Insights into the Factors Influencing Residents’ Travel Distance Distribution

Understanding intra-urban travel patterns through quantitative analysis is crucial for effective urban planning and transportation management. In previous studies, a range of distribution functions were modeled to lay the groundwork for human mobility research. However, few studies have explored the nonlinear relationships between travel distance patterns and environmental factors. Using travel distance data from ride-hailing services, this research divides a study area into 1 × 1 km grid cells, modeling the best travel distance distribution and calculating the coefficients of each grid. A machine learning framework (Extreme Gradient Boosting combined with Shapley Additive Explanations) is introduced to interpret the factors influencing these distributions. Our results emphasize that the travel distance of human movement tends to follow a log-normal distribution and exhibits spatial heterogeneity. Key factors affecting travel distance distributions include the distance to the city center, bus station density, land use entropy, and the density of companies. Most environmental variables exhibit nonlinear and threshold effects on the log-normal distribution coefficients. These findings significantly advance our understanding of ride-hailing travel patterns and offer valuable insights into the spatial dynamics of human mobility.