Research On The Nonlinear Effects Of Built Environment On E-scooter Travel Demand

The emergence of shared electric scooters(ESS),an emerging transportation mode,has already brought about dramatic changes to the transportation system.As of 2019,ESS have surpassed other micro-mobility vehicles to become the most used micro-mobility vehicles in the United States.As ESS services become more popular in the United States and the ESS market continues to grow,so does the research on these services.An unresolved issue is the nonlinear relationship between built environment and ESS usage,how the so-called "D" elements(i.e.density,diversity,design,accessibility)affect in a nonlinear way on ESS usage.This paper aims to use the ridership data of ESS in Los Angeles from April 2019 to February 2020 and related socioeconomic and built environment data to study and analyze the impact of the built environment and socioeconomic variables on ESS usage.The main contents are as follows: First of all,summarize the travel characteristics of ESS,the impact of the built environment on ESS usage,and the nonlinear impact of the built environment on it;Secondly,describe the data and data analysis used in this paper process,and conduct preliminary spatiotemporal feature analysis on travel data;Then,introduce the basic concepts of the machine learning model used in this paper and the principles of the three algorithms of gradient boosting decision tree(GBDT),XGBoost and random forest;In the end,by constructing GBDT,XGBoost and random forest model respectively to explore the nonlinear and threshold effects of built environment variables on ESS and the importance of each variable and draw three types of partial dependence plots(PDP),individual conditional expectation map(ICE)and accumulated local effect(ALE).Formally demonstrate the nonlinear effects of built environment variables on the ridership density of ESS.The research results show that the fitting results of the GBDT model are better among the three models.Regarding nonlinear effects,in addition,almost all variables have nonlinear relationships with ESS ridership density and have threshold effects.Regarding the ranking of feature importance,the top ten variables in the three models have strong commonality.Overall,restaurant density,road density,intersection density,mass transit station density,park density,population density,employment density,and parking lot density are the built environment variables that are more important,and the built environment variables are much more important on socioeconomic variables.This paper hopes to make transportation planning departments pay more attention to the link between the built environment and urban transportation by studying the nonlinear impact of the built environment on ESS travel.The research results provide a scientific theoretical basis for the formulation of measures such as optimizing the urban slow-moving system and promoting the development of shared transportation.