Inferring OD-Pairs and Utility-Based Travel Preferences for Shared Mobility System Users in a Multi-Modal Environment

This thesis develops new methods to identify individual traveler's preferences and at the same time, infer true" Origin-Destination (OD) locations based on incomplete route information of shared mobility system uses in a multimodal travel environment. Based on observations of traveler's route choices of a bike sharing system under various price setting, the devised method performs probabilistic reasoning to infer traveler's OD and preference. The Selective Set Expectation Maximization approach developed in this thesis is used to infer the unknown distribution of traveler preference vectors which are not mutually exclusive; treating the traveler's Origin-Destination as a latent variable. The unknown model parameters are found by maximizing the expected value of the log likelihood function with respect to the conditional distribution of the OD pairs under the observed bike segment usage in an iterative manner. The experimental results using simulation data demonstrate the accurate learning or inference of the true underlying distribution using the developed Selective Set Expectation Maximization algorithm and then inferring the true OD pairs using this learned distribution.