A1538
Title: Understanding bicycle route choice through explainable AI
Authors: Katrin Lubashevsky - TUD Dresden University of Technology (Germany)
Stefan Huber - TUD Dresden University of Technology (Germany)
Sven Lissner - TUD Dresden University of Technology (Germany)
Iryna Okhrin - Dresden University for Technology (Germany) [presenting]
Abstract: Bicycle route choice is analyzed with a focus on methodological comparison and explainable artificial intelligence (XAI). Using data from multiple German cities, detailed route-level attributes (e.g., infrastructure quality, traffic conditions, elevation) are combined with city-level characteristics (e.g., urban structure, cycling culture, environmental factors) to capture both local and contextual influences. A classical discrete-choice model (conditional logistic regression) is compared with machine learning methods (e.g., random forests, XGBoost, support vector machines, deep learning models). The emphasis is on the trade-off between predictive performance and interpretability. To address this, XAI techniques are applied to systematically evaluate feature importance and model behaviour across methods. Results indicate that machine learning models generally achieve higher predictive accuracy, while XAI enables meaningful interpretation despite their complexity. In contrast, conditional logistic regression provides transparent, theory-consistent insights but may be less predictive. Overall, the analysis demonstrates how combining statistical models, machine learning, and XAI yields a comprehensive understanding of cyclist preferences, supporting data-driven urban planning and the design of sustainable cycling infrastructure.
