Introducing Frameworks to Analyze Human Mobility Behavior with Advanced Computational Algorithms and Machine Learning Methods Using Mobile Device Location Data

Abstract

The emergence of mobile device location data (MDLD) provides new opportunities to analyze human mobility behaviors. The large penetration rate and the possibility of observing human mobility behaviors continuously are among the most important features of the passively collected mobile device location data. However, to utilize MDLD in mobility behavior analysis, comprehensive computational algorithms need to be developed to carefully process the data.This research proposes novel sets of frameworks to extract mobility context from the raw MDLD. First, this study introduces a set of algorithms to construct the travel behavior of mobile device owners along with the non-observable attributes of both trips and travelers by extracting trips, identifying significant activity locations of the travelers such as their home and work locations, and imputing the travel mode. The proposed algorithms in this study were tested against the state-of-practice and state-of-art algorithms developed in the literature. The proposed algorithms were shown to have superior performance compared to other methods. Next, this study further examines the usefulness of the proposed framework in providing near real-time insights on the evolution of human mobility behavior during the Coronavirus disease 2019 (COVID-19) pandemic. As a part of this study, a new metric has also been introduced to measure the social distancing practices from the mobility perspective. Additional investigations are also conducted to understand the linkage between the outbreak of COVID-19 and the mobility behavior of the communities. Lastly, this study seeks to develop a framework to investigate the evacuation behavior of individuals during a natural disaster and construct the evacuation evolution patterns and decisions based on the MDLD. This dissertation evaluates the importance of the historical mobility behavior of the device owners in their decision-making procedure during natural disasters using statistical discrete choice models.