Where can automated mobility-on-demand service thrive: A combined method of latent class choice and random forest

Transportation systems have reached a tipping point with autonomous driving technologies and innovative mobility services. There are a growing number of studies related to factors affecting the preference for automated mobility-on-demand mobility (AMoD) services. However, the various factors affecting the heterogeneity of autonomous driving services have not been thoroughly explored. Therefore, this study aimed to scrutinize the factors affecting the taste heterogeneity of autonomous driving services. A combined method of a latent class model and a random forest was suggested to overcome the overparameterization problem in a latent class model. The proposed model allows us to consider a wide range of variables affecting mode preferences at once, while it is challenging to interpret the direction and the magnitude of impact or heterogeneity compared to the latent class model. Therefore, the Shapley additive explanations (SHAP) was employed to interpret the results of the random forest. To investigate a variety of factors, two datasets, which are data including stated choice preferences, demographics, and attitudinal indicators from online questionnaire and regional characteristics from Statistics Korea, were combined based on the residential addresses of respondents. The latent class model revealed six classes based on transport mode preference. This suggests that there is taste heterogeneity in the preference for AMoD. SHAP analysis identified the magnitude and direction of the impact of the factors influencing taste heterogeneity for AMoD. The findings of this study can contribute to establishing transportation policies for autonomous vehicle diffusion and selection of pilot districts.