Relationship Between Built Environment of Metro Station and Passenger Attraction Considering Spatial Heterogeneity
Machine learning models have been extensively applied in exploring the interaction between the built environment and passenger flow.However,machine learning primarily considers global relationships and fails to capture spatial variations.To address this issue,this paper defined 11 built environment variables from the aspects of density,diversity,design,destination accessibility,availability,and network connectivity.The study proposed an integrated ensemble analysis model,SLightGBM,combining Light Gradient Boosting Machine(LightGBM)with Geographically Weighted Regression(GWR),to investigate the spatial heterogeneity and nonlinear impact of the built environment on the attractiveness of station coverage.The SLightGBM model was compared with the LightGBM,Ordinary Least Squares(OLS),and GWR to demonstrate its regression superiority.The results from Xi'an city indicate that:(1)The SLightGBM model showed better performance than other models,with R2 value of 0.68,MAE of 8379.16,and RMSE of 11797.19.(2)The factors of the built environment vary across spaces.The densities of the employment and bus stops are most important in central areas,whereas the density of restaurants is more prominent in the southern regions.(3)Higher employment and restaurants densities are positively correlated with the attractiveness of metro ridership,while the minimum transfer times are negatively correlated with the attractiveness of metro ridership,showing a strong combined effect.This study indicates the importance of understanding spatial differences and threshold effects of these factors in urban planning and public transport system improvement.
NETL
NSTL
万方数据
