Exploring the relationships between usage efficiency for shared e-scooter services and urban built environment, socio-demographic factors: A machine learning approach

Abstract

Shared e-scooters are sometimes presented as a sustainable mobility option, but their environmental impact depends on their utilization. When utilization is low, idle e-scooters increase operational costs and decrease overall energy efficiency. This thesis explores whether neighborhood-level built environment and socio-demographic characteristics influence the usage efficiency of shared e-scooters in Stockholm, Sweden. Usage efficiency is measured using time-to-book (TTB), defined as the time an e-scooter stays idle between trips.
E-scooter trips from June 2021, Statistics Sweden (SCB) demographic data and OpenStreetMap points of interest (POIs) were aggregated at the demographical statistical area (DeSo) level. Variables included POI densities, age brackets, education, gender ratio, car ownership, housing situation and income categories. 229 DeSo areas were included, and both multiple linear regression (MLR) and random forest (RF) models were applied to evaluate predictive performance and identify key influencing factors. The RF model outperformed MLR, capturing non-linear relationships and providing higher predictive accuracy.
The results show that built environment characteristics have a stronger and more consistent effect on e-scooter usage efficiency than socio-demographic attributes, although both contribute. The weak influence of some usually strong predictors like age and gender may be due to the limitations of the aggregated data or usage by tourists and other groups not captured in the residential statistics. Spatial aggregation at DeSo level and the absence of user level demographic information were notable constraints and may mask some finer scale behaviors. Despite the limitations, the results show that machine learning produces better results when modelling e-scooter efficiency.

Popular Abstract

Cities around the world are trying to become greener and more efficient, and one way they’re doing that is by promoting shared e-scooters. These battery-powered scooters are touted as a greener alternative to car use, helping reduce carbon emissions. However, that can be true only if people actually use them. In many cities, e-scooters spend a lot of time just sitting around unused, which wastes energy and undercuts their environmental benefits.
This thesis looked into what makes some neighborhoods better than others in e-scooters usage. I examined Stockholm, Sweden to see if the characteristics of a neighborhood like its layout or the kinds of people who live there help predict how often e-scooters are used? To find out, I combined data on e-scooter trips with information about local demographics and features like restaurants, schools, and food places. Then I used machine learning, specifically a method called random forest, to analyze the data and compare it with linear regression - a traditional statistical method.
As it turns out, areas with lots of places to eat, like cafes and restaurants, had the most efficient e-scooter usage. Wealthier neighborhoods also saw more frequent use. On the other hand, places with higher levels of government assistance saw scooters sitting idle for longer periods of time. Interestingly, age and gender patterns didn’t always match assumptions - older groups used scooters more than expected in some areas, and young adults were not always the dominant users.
Overall, machine learning gave a better picture than traditional methods. It could account for patterns and non-obvious connections that older models might disregard. These insights could be useful also for city planners and transportation companies. By understanding where and why e-scooters are being used, the fleet can be better repositioned and their usage more effective. This adjustment could in turn help make urban transportation greener and more responsive to the needs of city residents.

• author: Gunka, Magdalena
• supervisor: Pengxiang Zhao
• organization: Dept of Physical Geography and Ecosystem Science
• course: NGEM01 20251
• year: 2025
• type: H2 - Master's Degree (Two Years)
• subject: Earth and Environmental Sciences
• keywords: Physical Geography and Ecosystem analysis, Geography, Micromobility, E-scooter utilization, Random forest, Socio-demographic analysis, Urban transportation
• publication/series: Student thesis series INES
• report number: 749
• language: English
• id: 9215191
• date added to LUP: 2025-11-11 09:24:38
• date last changed: 2025-11-11 09:24:38