Thesis Title: Bayesian Nonparametric Regression through Mixture Models

This thesis studies Bayesian nonparametric regression through mixture models. These types of models are highly flexible, yet also numerous, which raises the question of how to choose among the models for the application at hand. In answer to this question, we derive predictive equations for the conditional mean and density and carefully analyse the quantities involved. Our main contributions to the subject are a detailed study of the predictive performance of existing models, the identification of potential sources of improvement in prediction, and the development of novel procedures to improve prediction. The models developed are applied in three studies of Alzheimer’s disease, with the aim of diagnosis of the disease based on AD biomarkers and investigation into the dynamics of AD biomarkers with increasing age.