How Do Machine Learning Approaches Measure Up? A Comparative Evaluation of Machine Learning Techniques for Bicycle Route Choice Analysis and Modelling.

Analyzing cycling behavior holds significant relevance in transport planning and modelling, as it directly impacts the design process and optimization of urban bicycle networks. Several studies have extensively investigated bicycle route choice using GPS data, shedding light on the preferences and behaviors of cyclists. Although machine learning methods are increasingly used in transport planning, there remains a notable gap in research applying machine learning methods to the analysis and modelling of bicycle route choice. Thus, the suitability of machine learning for this purpose remains uncertain. The aim of this study is to assess the utility of machine learning approaches in comparison to established methods, such as the logistic regression model. Therefore, a comprehensive GPS dataset comprising 25,730 GPS tracks from 1,361 cyclists in Dresden (Germany) is used. The results indicate that machine learning methods perform comparably well to the traditional logit model in terms of model quality. However, the findings also show that the strength of the effects of influencing factors varies across different machine learning models and unlike the established logit model, machine learning models still lack transparency regarding the interpretation of individual factors.