Abstract
My dissertation includes three projects that I worked on in collaboration with my committee members and other students and faculty. Each of these projects facilitates model selection and model evaluation through an experimental design, a modeling framework that respects the structure of compositional data, and a measure of the impact of prior information on the fitted model. Chapter 2 presents an experiment design method for model selection and checking called SeqMED. Chapter 3 introduces a new modeling framework for selecting predictive signatures from compositional data. In Chapter 4, I introduce a measure that quantifies the impact of the kernel choice on Gaussian Process regression.
Pantoja, Kristyn Jae (2022). Methods for Model Discrimination and Evaluation: An Experimental Design, a Modeling Framework for Microbiome Data, and a Measure of Prior Impact. Doctoral dissertation, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /198034.