Predicting cycling flows in cities without cycling data

Abstract

Cycling is a potential tool to mitigate many of the problems faced by urban populations today. Encouraging the use of bicycles as a legitimate mobility tool, however, demands adequate knowledge of current mobility patterns, such as locations of trip generation and attraction. Unfortunately, cities usually do not gather enough data to adequately understand cycling demand. We propose models based on spatial econometrics and gradient boosted regression trees which can be trained with data from cities with mature cycling cultures and then applied to cities still in their cycling infancy to supply city officials with a better estimate of potential future OD matrices. We perform a case study in the Boston Metropolitan Area and show results comparing both types of models.