Estimating factors influencing mode choice behaviors has posed a significant challenge for decision makers. In this paper, we propose a three-step method to estimate mode choice between public transportation and private vehicles within a heterogeneous metropolitan area. First, we implement a deep learning model, Altaïr, capable of inferring travel times and travel flows by leveraging multi-source input data (Step 1). To identify homogeneous sub-regions regarding mode choice behaviors, two data clustering models are performed: k-means and a Gaussian Mixture Model (GMM, Step 2). The GMM reveals three spatial clusters based on the relationships between relative travel times and relative travel flows in public transportation and private vehicles. Moreover, an econometric model (robust ordinary least squares) is employed to identify additional explanatory variables, including sociodemographic features and location variables (Step 3). This hybrid method is currently employed in the Paris metropolitan area (France). At the metropolitan level, we find that competitive travel times in public transportation lead to higher ridership. Conversely, when the time ratio exceeds approximately 3.5–4, public transportation use becomes negligible in comparison to private vehicles. The method's results align with those of the regional Household Travel Surveys (HTS) regarding the individual attributes influencing public transportation use (share of women and students in the municipality of residence) or private vehicle use (share of car owners and people over 55). There is significant car use in peripheral areas, but public transportation accessibility promotes higher flows in this mode everywhere, with the exception of central Paris. • We use a three-step method to estimate mode choice behaviors in a metropolitan area. • A deep learning model infers travel times/ travel flows from multi-source input data. • Using the relationship between the two variables, a GMM reveals spatial clusters. • An econometric model highlights other explanatory variables for mode choice behaviors. • Our method can guide transport policies in areas where HTS are too difficult or costly to implement.
Boënnec et al. (Sun,) studied this question.