To what extent walking and biking are substitutes or complements to public transport? Interpretable machine learning findings from the University of Lyon, France

This study examines the dynamic relationship between active mobility and public transport among university students, focusing on how this interaction varies based on home-campus distance. Using sequential and randomized tree-based ensemble machine learning models and interpretation techniques on survey data, we uncover nuanced patterns in behavior regarding the choice of transport modes for commuting. Key findings are that biking complements public transport for distances less than two kilometers and greater than ten kilometers, and walking complements it for distances over two kilometers, while some subgroups of students substitute walking for public transport in short distances. Biking and public transportation are observed to be substitutes for each other in distances between approximately two and ten kilometers. Results also suggest a substitution effect between walking and biking for short distances. Finally, we identify distinct walking and biking clusters around certain campuses. Our study highlights the potential of policies promoting active mobility to reduce motorized transport use among students, thereby mitigating social, environmental, and health risks in urban settings.