Applying machine learning approaches to model travel choice between micro-mobility services

Abstract: Shared micro-mobility gradually becomes a crucial part of human daily transportation. To develop the shared micro-mobility, discovering the important influence factors of each travel mode is a key aspect. However, there are scarce studies that adopt machine learning methods to model travel choice between shared micro-mobility services and identify the crucial determinants for each mode. This study aims to apply four different machine learning methods (random forest, support vector machine, artificial neural network, and logistic regression) to simulate the decision schema in time and space and examine the factors that significantly influence citizens’ travel mode choice in Zurich, Switzerland. Collected data are used to build the choice set, which mainly includes trip attributes and external environment factors. Then, the four Machine Learning methods are used to examine the importance of each influence factor by permutation importance, and the performances of four ML models are compared. How the top 6 influence factors with higher importance affect the human choice of shared micro-mobility service is analyzed as follows. The Random Forest model showed the best-predicted performance. With respect to the feature importance in the RF model, trip attributes, such as the duration and the length of the trip, are identified as the most important influence factor, followed by some POI type and density around the destination, like public facility, and education services. By contrast, weather affects people’s choices slightly. It is noteworthy to see that dockless facilities always have priority when there is the same type of docked facilities without considering other variables, but docked services are still preferable under some situations. The results are valuable to policymakers and shared services provided by companies to adjust the shared micro-mobility system and contribute to better sustainable transportation.