The nonlinear relationship between built environment and cycling propensity for different travel purposes - based on extreme gradient boosting decision tree.

Bicycles, as a flexible, convenient, and low-carbon mode of transportation, possess significant advantages in short-distance travel. Linear and generalized linear models have been widely employed for modeling in previous studies, while the nonlinear relationship between the built environment and cycling behavior has received scant attention. Consequently, this paper delves into the significance of individual and family socio-economic attributes, travel characteristics, and built environment attributes regarding bicycle choice behavior, as well as the nonlinear relationship and threshold effect between the principal variable and the propensity for bicycle choice for two distinct travel purposes, based on the data from the 2018 Daily Trip Survey in Xianyang, China. The results reveal that: (1) The prediction accuracies of the cycling choice models for compulsory and discretionary trips were 89% and 80%, respectively. (2) The built environment exerts a more substantial impact on the two travel purposes, with the impact of compulsory trips being notably greater than that of discretionary trips. The relative importance of variables encompasses the proportion of non-vehicle isolation belts, the non-motor lane parking ratio, population density, distance from the center, and POI diversity, among others. (3) In the two travel activities, the relationships between the main variables and threshold effects vary. These findings offer policy implications for implementing differential interventions to promote bicycle usage in cities.