Using machine learning for direct demand modeling of ridesourcing services in Chicago

The exponential growth of ridesourcing services has been disrupting the transportation sector and changing how people travel. As ridesourcing continues to grow in popularity, being able to accurately predict the demand for it is essential for effective land-use and transportation planning and policymaking. Using recently released trip-level ridesourcing data in Chicago along with a range of variables obtained from publicly available data sources, we applied random forest, a widely-applied machine learning technique, to estimate a zone-to-zone (census tract) direct demand model for ridesourcing services. Compared to the traditional multiplicative models, the random forest model had a better model fit and achieved much higher predictive accuracy. We found that socioeconomic and demographic variables collectively contributed the most (about 50%) to the predictive power of the random forest model. Travel impedance, the built-environment characteristics, and the transit-supply-related variables are also indispensable in ridesourcing demand prediction.