Modeling Cycling Route Choice Using Convolutional Neural Networks

Abstract

Bicycles offer significant potential for sustainable urban transport by reducing emissions and noise when replacing motorized traffic. Understanding cyclists’ route choices is crucial for city planners, and traditionally is analyzed using multinomial logit (MNL) models. Recent advancements in machine learning (ML) show capability for greater predictive accuracy. This study explores potential of the ML modeling technique convolutional neural networks (CNN) for bicycle route choice modeling using GPS data from around 183,000 trips across six German cities. By comparing CNN with MNL, we evaluate their strengths, limitations, and implications for cycling infrastructure planning.