A0470
Title: Analyzing the route choice of cyclists using machine learning models
Authors: Katrin Lubashevsky - TUD Dresden University of Technology (Germany) [presenting]
Iryna Okhrin - Dresden University for Technology (Germany)
Stefan Huber - TUD Dresden University of Technology (Germany)
Sven Lissner - TUD Dresden University of Technology (Germany)
Abstract: Cycling is a crucial part of the transition to a more climate-friendly transportation system. It is therefore advisable to promote cycling in transport planning, where an understanding of the influences on the choice of cycling route is necessary for efficient planning. Using GPS tracks of cycling trips and additional route information, the route choice can be modeled in various ways. To date, logit models have been the predominant type of models used in this domain. The present contribution employs a variety of machine learning techniques, including neural networks, support vector machines, decision trees, and random forests, to address the route choice problem. Data from the Germany-wide "City Cycling" campaign is used for this purpose, with the city of Freiburg, in particular, being the subject. A total of 418,620 bicycle trips were recorded for the city of Freiburg. The resulting models are evaluated and compared according to their interpretability (using partial dependence plots and variable importance plots) and predictive quality (using metrics such as recall, accuracy, and F1-measure). First, results have already shown that some of these methods show higher prediction quality than the classical logit models.
