Thesis Defense: Qi Wang, “System Dynamics and Machine Learning Techniques for Studying Resilience in Public Health”

December 10, 2020





Doctoral Candidate


Tuesday, December 22

11:00 AM

Contact Elena Shichkova for access to this presentation.

“System Dynamics and Machine Learning Techniques for Studying Resilience in Public Health”

Systems Dynamics (SD) and Machine Learning (ML) are analytical methods that are becoming more broadly applied to studies in public health. This dissertation focuses on public health aspects of resilience, with an emphasis on distressed populations. We analyze a broad array of topics, including community resilience to natural disasters, suicide prevention interventions, cognitive resilience to aging effects, and interventions to mitigate the impacts of community violence on children.  We also include SD and ML studies in early childhood educational investment and a Bayesian analysis of wiki-surveys.

We begin with the development and calibration of a system dynamics model of community resilience, COPEWELL. We propose a stepwise approach to developing a measure set that combines domain expertise with statistical analyses. For our second study, we use SD to analyze processes of suicide prevention for refugees living in migration camps in Thailand to understand their causes, consequences, and mechanisms. An SD model is used to show the different impacts of proposed interventions on rates of suicide ideation and attempts. In our third study, we explore a novel application of supervised ML to study time-varying processes of cognitive reserve and resilience in an aging population in Baltimore City. For our fourth study, we worked with researchers at the United Stated Agency for International Development (USAID) and World Vision International to build an SD model of community violence in gang-controlled urban neighborhoods in Honduras and El Salvador. The long-term goal in this collaborative effort is to design interventions that would promote child protection against violence. An interactive webpage was developed for stakeholders to visualize the model and explore influential factors. In our fifth study, we use reinforcement learning to model parental investment in early childhood development. We integrate this model with an economics-based model of children human capital formation. We use Bayesian updating to assess the future state of the stock of human capital. In our sixth study, we construct and calibrate an SD model to examine the influence of community factors on suicide risk. This study lays the foundation for a larger project to examine the impact on community-based support programs such as home visitation programs on the rates of suicidal behaviors. In the final application of this thesis, we demonstrate a scoring and ranking approach for analyzing the results of wiki-survey voting using a maximum a posteriori probability (MAP) estimate. Two case studies are examined to test the accuracy of the estimating procedure.

This thesis demonstrates the broad applicability of system dynamics and machine learning methods on high-priority public health research topics. Mathematical details and software code are provided in the appendix for researchers interested in this growing direction of research.


Back to top