Urban cycling flow prediction using machine learning

Abstract

Promoting bicycle use in cities is an integral part of public policies aiming to promote the use of non-polluting, active means of transport, which are beneficial to the health of the population by counteracting a sedentary lifestyle. Choosing where to build bicycle lanes is of paramount importance during this process. In this project we will explore data related to cycling trips in cities with mature bike-sharing systems. We will use machine learning models to infer mobility flows from socioeconomic data, points of interest, topology, and more, in multiple cities. Our approach will be to use transfer learning methods to tune these models so they can be used in cities without bike-sharing systems, or with systems that are still in their infancy, allow for better planning when choosing where the bike-sharing stations should be located and where to build new cycling infrastructure, such as bicycle lanes. (AU)