Abstract
Global development is a comprehensive package that deals with critical issues in society, the economy, environment, and government. It is a combination of strategy and techniques that have to deal with the intertwined problems such as the lack of financial services, poor health care, racial segregation, and unsanitary environment. In order to design an effective development project, a concrete understanding of such issues is necessary, which requires embracing the complexity that conventional development models have typically ignored. At the same time, reliable information that supports causal discovery and impact evaluation is an inevitable ingredient to understand in global development practice. However, high-level information such as national statistics and archival records is exorbitantly expensive to acquire, if not impossible, in developing countries.
The purpose of this dissertation is to propose methods that utilize unstructured data with two engineering techniques: machine learning and simulation. As new sources of unstructured data proliferate, such as satellite images, text, and transaction records, there are exciting possibilities that such data coupled with machine learning could provide high-level information. At the same time, simulation techniques provide a mechanism that models causal relationships in complex systems. This dissertation frames three broad approaches to incorporate machine learning and simulation, jointly and separately, in the context of global development, while also showcasing five applications in multiple sub-systems in international development: social, economic, governance, and environmental systems. These unstructured data-driven methods can provide compelling approaches that go beyond just aggregating data or averaging out superficial behaviors.
