Characterizing Shared Mobility Operator and User Behavior Using Big Data Analytics and Machine Learning 335 views

Towards more sustainable use of resources in cities, there is a rising trend in shared mobility for collaborative consumption. As a condition of working with cities, third party organizations managing shared vehicle fleets often have to provide public access to real-time data describing the location of vehicles. These datasets hold enormous value for monitoring and evaluating emerging transportation services; however, a major challenge for city planners and regulators remains extracting the value from streaming transportation data by leveraging analysis and visualization methods. E-scooters are an emerging shared mobility service that have been adopted in cities across the world, but, despite their popularity, cities are still searching for more effective monitoring methods in order to understand the benefits brought to their communities or lack thereof. Using real-time e-scooter data from Charlottesville, Virginia as a case study, this work aims to characterize operator and user behavior by using big data analytics and machine learning to gather important insights. Specifically, this work provides the following contributions via three analytical studies: (1) Study I demonstrates how e-scooter data can be harvested from streaming GPS traces and then aggregated and spatially joined with demographic, employment, and built environment data. A multiple regression analysis examining the relationships between these datasets revealed that e-scooter distribution was influenced by economic activity whereas e-scooter use was influenced by micro-transit need factors and built environment characteristics. (2) Study II presents data aggregation and visualization approaches for monitoring and evaluating e-scooter operator distribution decisions, showing that utilization is a suitable measure for planning and revealing that there is room for improvement for equitable fleet distribution. (3) Study III shows the efficacy of using Latent Dirichlet Allocation to characterize user trip behavior from an unstructured set of estimated e-scooter trips. Findings suggest that trip behavior differed significantly during periods with increased student population influxes. Charlottesville planners and regulators may use the results and methods presented in this work to make data-driven decisions for improving micro-mobility as a service for the community they serve.

MS (Master of Science)

machine learning; data visualization; data fusion; shared mobility; micro-mobility; big data analytics; multiple regression; Latent Dirichlet Allocation; geospatial data; real-time data

English

All rights reserved (no additional license for public reuse)

2021-04-27