A Generative Approach for Zonal Inference and Transportation Modelling

Confirmation seminar presented by Eren Arkangil.

Transportation modeling and traffic analysis have traditionally followed the sequential four-step model procedure. However, an alternative approach known as activity-based models is proposed, wherein travel demand is derived from micro-level activities of individuals. The success of activity-based models and simulations is greatly dependent on the quality of the dataset used. Typically, Census or Household Travel Survey datasets are used due to their inclusion of rich socioeconomic information. However, the publicly available sample of these datasets only represents a small and biased proportion of the population. Consequently, it is crucial to develop a data-driven model that can accurately capture the marginal and joint distributions of the actual population for reliable analysis. The generative approach aims to accurately predict these distributions while addressing the lack of publicly available data by up-sampling the small sampled datasets.