An interpretable machine-learning mode choice model incorporating bicycle route quality

The attractiveness of bicycle routes can strongly influence mode choice. As cities promote sustainable and active modes of transport, accounting for bicycle route quality becomes increasingly relevant in mode choice models used for transport planning, policy evaluation or traffic simulations. We present a data-driven mode choice model that integrates characteristics of bicycle routes generated using a multicriteria bicycle routing model. The bicycle routing and machine-learning mode choice model are both developed using revealed-preference data from GPS-tracking-based travel diaries. The eXtreme Gradient Boosting mode choice model captures the impact of different types of bicycle infrastructure, and the accuracies are improved overall. We assess model plausibility by performing a SHAP interpretability analysis and simulated infrastructural intervention. By accounting for a wide range of factors that affect bicycle route quality, it is possible to better model and facilitate the shift toward sustainable transportation.