A COMPARISON OF FREQUENTIST AND BAYESIAN APPROACHES FOR LINEAR GAUSSIAN PROCESS MODELS

Open Access
Author:
Atiyat, Muhammad
Graduate Program:
Statistics
Degree:
Master of Science
Document Type:
Master Thesis
Date of Defense:
None
Committee Members:
  • Debashis Ghosh, Thesis Advisor
  • Murali Haran, Thesis Advisor
Keywords:
  • gaussian process model
  • spatial models
Abstract:
Spatial data (data that are geographically referenced) are commonly encountered in varied fields such as ecology, epidemiology, public health, and geoscience. We consider practical issues with using linear Gaussian process models, which are among the most popular models for analyzing spatial data. We summarize some commonly used frequentist and Bayesian approaches for modeling spatial data via Gaussian processes. In the Bayesian context we review some standard approaches for selecting appropriate priors. We also compare estimation and prediction for Gaussian process models via a simulation study and through an application of our methods to a spatial data set used for studying crop epidemics. We conclude with some practical recommendations based on our study.