Applications of Bayesian inference for the origin destination matrix problem

Official URL: http://risc01.sabanciuniv.edu/record=b1669821 (Table of Contents)

This thesis presents a study of estimating the probability matrix of an origin-destination model associated with a two-way transportation line with the help of Bayesian inference and Markov chain Monte Carlo methods, more specifically, Metropolis within Gibbs algorithm. Collecting the exact count data of a transportation system is often not possible due to technical insufficiencies or data privacy issues. This thesis concentrates on the utilization of Markov chain Monte Carlo Methods for two origin-destination problems: one that assumes missing departure data and one that assumes the availability of differentially private data instead of the complete data. Different models are formulated for those two data conditions that are under study. The experiments are conducted with synthetically generated data and the performance of each model under these conditions were measured. It has been concluded that MCMC methods can be useful for effectively estimating the probability matrix of certain OD problems.