Zero-inflated Models for Semi-continuous Transportation Data

Abstract

Zero-inflated models have been widely studied and are commonly used in the transportation safety area. Despite the success of zero-inflated models to analyze static data with counting outcomes, challenges remain in the examination of zero-inflated data with semi-continuous and auto-correlated longitudinal outcomes. This dissertation aims to explore different approaches to tackle the existing challenges in semi-continuous zero-inflated data analysis. The dissertation begins with a discussion of challenges with existing zero-inflated models in modelling semi-continuous data and time-series data. Then, our recent works on a variable selection method for semi-continuous zero-inflated models and a dynamic model for semi-continuous zero-inflated time-series data are presented. Simulated data was used to validate the proposed models. And finally, we demonstrate the proposed models using two different transportation datasets. These datasets are from a driving simulator study and a field operational test. The results suggest that the proposed models can capture the differences in driving behavior between individuals and between different driving situations, which have implications for the design of in-vehicle assistance systems.