Built environment impacts on zonal shared e-scooter expenses: A Bayesian learning approach

Shared e-scooters are reshaping urban mobility, yet trip expense patterns, a key to operator viability, remain unexplored. This study examines how built environment factors affect zonal-level shared e-scooter trip expenses in Chicago, using a novel lognormal regression model enhanced by Bayesian Additive Regression Trees (LN + BART). The model outperforms traditional methods by accommodating the right-skewed distribution and capturing the nonlinear effects on the trip expenses. Results reveal threshold effects: areas with higher median income level, higher POI (Point of Interest) density, and closer distance to CBD (Central Business District) yield disproportionately higher revenues. However, zones with higher percentages of car-free households show lower e-scooter usage, highlighting affordability barriers despite clear mobility needs. This research advances transport economics by combining distribution-aware modeling with Bayesian machine learning, enhancing prediction and interpretability. It also offers important insights for operators to optimize deployment.