Comparing built environment effects on bike-sharing and electric bike-sharing usage: a spatiotemporal machine learning approach

Shared micromobility has been widely recognized as a promising solution for promoting sustainable urban transportation, experiencing rapid growth and diversifying into various services, such as bike-sharing (BS) and electric bike-sharing (EBS). However, existing studies have primarily examined BS and EBS separately, leaving comparative analyses of their travel patterns and determinants notably limited. Moreover, although machine learning approaches have become prevalent for modeling nonlinear relationships, these methods typically overlook spatiotemporal heterogeneity, potentially resulting in biased estimations and inaccurate interpretations. To address these gaps, this study develops a novel modeling framework integrating XGBoost with geographically and temporally weighted regression (GTWR), enabling simultaneous consideration of spatiotemporal heterogeneity and nonlinearity. Using trip data from Hefei, China, we comparatively analyze the travel characteristics of BS and EBS and apply the integrated modeling framework to investigate the built environment’s influence on both modes. The results indicate that both BS and EBS exhibit distinct peak-hour usage patterns, while spatially, BS usage is concentrated in downtown areas and EBS usage is more evenly distributed citywide. Among examined factors, distance to metro stations and employment density emerge as the most significant predictors for both modes. Additionally, nonlinear relationships reveal that higher branch road density and lower major road density are associated with increased BS but reduced EBS usage, while land use mix demonstrates clear threshold effects, beyond which usage of both modes significantly increases. These findings provide valuable insights for operators to optimize fleet deployment and for policymakers to design targeted interventions supporting coordinated and sustainable shared micromobility development.