Predicting Hourly Shared E-scooter Use in Chicago: A Machine Learning Approach

Abstract

Shared e-scooter programs were first implemented in 2017 to solve problems with the current transportation landscape. Combining ideas from mobility as a service, micromobility, and multimodal transportation, shared e-scooter systems and other forms shared of transportation programs have the potential to reduce or eliminate the need for unsustainable personal vehicles. However, shared e-scooters can create more problems than they solve. Some problems e-scooters can create include vandalism, lack of accessibility, hazards for the rider and pedestrians, and added pollution to the environment. With proper management, these problems can be mitigated. Using frameworks from optimizing bike sharing programs, a predictive algorithm for shared e-scooters to predict hourly trips for e-scooter pilots was created. Features that help predict hourly e-scooter trips include time of day, number of days since its inception, rainfall, wind speed, and more. Machine learning models with the best accuracy at predicting e-scooter trips includes K nearest neighbors, decision tree, and random forest. Shared e-scooter system managers can use these models for optimal allocation of e-scooters to maximize ridership.