Exploring usage patterns and determinants of shared E-scooters using data-driven methods: A study of Gothenburg, Sweden

Abstract

Understanding the usage patterns of shared e-scooter is crucial for decision-makers to implement effective policies and for e-scooter operators to optimize their services. This study aims to investigate the spatial and temporal usage patterns of e-scooters in Gothenburg, Sweden, and to analyze the impact of various factors, such as socio-economic and built environment factors, on e-scooter demand. Analysis of the data is done by employing data-driven methods and machine learning models, including hierarchical clustering, XGBoost, and random forest. The analysis consisted of two major parts: the usage pattern analysis, which primarily used e-scooter transaction data to analyze demand and trip patterns, and the influencing factors analysis, which analyzed several factors to determine their influence on e-scooter demand. The findings reveal distinct demand patterns across different areas and time slots in Gothenburg, with the highest average trip duration occurring during nighttime on weekends. The analysis considers 35 variables, categorized into temporal and spatial factors. Bus stops emerge as the most significant spatial determinant, followed by public buildings, commercial buildings, residential roads, other roads, and health POIs. Overall, built environment factors seem to impact e-scooter demand substantially more than socio-economic factors. Among the temporal determinants, it is the month that has the most significant influence, followed by minimum temperature, average precipitation, and weekday. A partial dependency analysis further explains the relationship between each determinant and e-scooter demand. The findings of this study provide valuable insights for e-scooter operators and policymakers to better understand and cater to the needs of e-scooter users in urban environments.