Enhancing Decision-making in Smart and Connected Communities with Digital Traces

Abstract

The ubiquitous use of information communication technologies (ICTs) enables generation of digital traces associated with human behaviors at unprecedented breadth, depth, and scale. Large-scale digital traces provide the potential to understand population behaviors automatically, including the characterization of how individuals interact with the physical environment. As a result, the use of digital traces generated by humans might mitigate some of the challenges associated to the use of surveys to understand human behaviors such as, high cost in collecting information, lack of quality real-time information, and hard to capture behavioral level information. In this dissertation, I study how to extract information from digital traces to characterize human behavior in the built environment; and how to use such information to enhance decision-making processes in the area of Smart and Connected Communities.

Specifically, I present three case studies that aim at using data-driven methods for decision-making in Smart and Connected Communities. First, I discuss data-driven methods for socioeconomic development with a focus on inference of socioeconomic maps with cell phone data. Second, I present data-driven methods for emergency preparedness and response, with a focus on understanding user needs in different communities with geotagged social media data. Third, I describe data-driven methods for migration studies, focusing on characterizing the post-migration behaviors of internal migrants with cell phone data. In these case studies, I present data-driven frameworks that integrate innovative behavior modeling approaches to help solve decision-making questions using digital traces. The explored methods enhance our understanding of how to model and explain population behavior patterns in different physical and socioeconomic contexts. The methods also have practical significance in terms of how decision-making can become cost-effective and efficient with the help of data-driven methods.