Data-driven insights into (E-)bike-sharing: mining a large-scale dataset on usage and urban characteristics: descriptive analysis and performance modeling

Abstract Bike-sharing is considered one of the most sustainable forms of urban transportation Abouelela (Trans. Res. Part A: Policy Pract. 169:103602, 2023) and, therefore, understanding what determines the efficiency of a bike-sharing system (BSS) is crucial. Various factors from different domains, like built environment or behavioral studies, have been demonstrated influential in single BSSs and small-scale datasets (from a single or few BSS). However, the multitude of potential factors requires rigorous analysis and application of statistical models on large datasets. Due to the unavailability of large-scale datasets comprising hundreds of BSSs and unified databases containing variables, interactions are only known to a limited extent. Here we show the usage patterns of different types of BSSs, and draw statistically valid conclusions on the relevance of factors from different domains by gathering over 43 Mio km open data from 267 BSSs and 108 predictors, as well as applying time-series clustering, building predictive classification models, and fitting a stepwise regressive Ordinary Least Squares (OLS) model. Results show that operational factors like operating time, design factors like operational area and share of electric bikes, sociodemographic factors like age and gender, built environment like cycling infrastructure, and economic factors like overnight stays all play an important role. Descriptive statistics reveal significant room for improvement in the operations e-bike-sharing system (electric BSS) and show how climate influences operational organizations. We anticipate this publication and the associated dataset, which we make openly available, to spark discussions about the design and operations of BSS and to be a starting point for further analyses.