Park & Ride (P&R) enables railway users to access the transit network by means of their own cars. Its usage at the regional level can be analyzed on the basis of Household Travel Surveys (HTS). To overcome the HTS limitations in sample size and refreshment, this paper is aimed to combine such HTS for learning the car2rail intermodality phenomenon of individual mobility, with an Automated Fare Collection (AFC) database for inferring it over a very large set of individual trips. The approach involves three steps: (i) the HTS-based featuring of Origin-Destination (O-D) trips; (ii) the treatment of the AFC dataset using the dynamic path search and ad-hoc rules based on General Transit Feed Specification (GTFS) data, to yield AFC rail O-D trips; and (iii) the supervised machine learning of P&R usage based on the HTS and AFC data, considering three methods (Support Vector Machine, Decision Tree and Artificial Neural Network). Application to the Paris – Ile-de-France region with 2010 HTS and 2019 AFC data revealed three types of intermodal trips by an unsupervised machine learning algorithm, two of them at morning peak hours with either short or long rail distances, and the last one after the evening peak.
展开▼
