Tracing the effects of COVID-19 on short and long bike-sharing trips using machine learning

Abstract

COVID-19 drastically changed human mobility, including bike-sharing usage. Existing studies found positive impacts of COVID-19 on bike-sharing use. However, their analysis focused on the first year of the COVID-19 pandemic. This study traces the effects of COVID-19 by including the bike-sharing data of the second and third years of the pandemic to provide more empirical evidence. We pre-defined short and long bike-sharing trips. Data collection and the effects of COVID-19 on both trips were separately addressed using public bike-sharing data in Seoul, South Korea. We conducted a time series and hot spot analysis to trace temporal and spatial bike-sharing usage changes. Our study applied a machine learning tool with Random Forest regression modeling to examine COVID-19 effects on two types of bike-sharing trips. Its impact is measured by looking at feature importance and calculating the SHapley Additive exPlanations (SHAP) value. The amount of bike-sharing usage continued to grow during the pandemic, with long bike-sharing trips being more prominent. A significant increase in the number of short bike-sharing trips was observed in the second year. Both short and long trips showed growth in the third year, even with a high number of COVID-19 cases reported. There were no significant seasonal changes in the spatial concentration of both trips. COVID-19 and the vaccination response positively impacted bike-sharing use in Seoul, highlighting our resilience in adapting to changes in human mobility.