An Unsupervised Learning-based Approach for User Mobility Analysis of E-Scooter Sharing Systems

Human mobility analysis is a key method for understanding urban dynamics and mobility optimization. Novel last-mile mobility, called micromo-bilities, that includes shared bicycles, electric bicycles (e-bikes), and electric scooters are seeing rapid widespread acceptance in major cities. Compared with existing mobility data such as cars, buses, and trains, the majority trip distance of micromobilities is short, typically less than a few miles. The riders use them for commuting, sightseeing, shopping, and/or fun. By using the mobility data of micromobilities, we can observe more fine-grained human mobility in urban areas than existing data sources. In this paper, we present an unsupervised learning-based technique to analyze human mobility in urban areas and to study user clusters for such micromobility services. In our approach, we cluster user mobility patterns by using non-negative tensor factorization (NTF) from area-based trip data (which only included locations of origin and destination). Our approach was applied to micromobility data collected from Chicago and Washington, D.C., and we observed characteristic patterns.