Theses and Dissertations

Advisor

Sepehrifar, Mohammad

Committee Member

Popescu, George

Committee Member

Zhang, Jialin

Committee Member

Woody, Jonathan

Committee Member

Zhou, Qian; Wu, Tung-Lung

Date of Degree

8-7-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Dissertation - Open Access

Major

Mathematical Sciences (Statistics)

Degree Name

Doctor of Philosophy (Ph.D.)

College

College of Arts and Sciences

Department

Department of Mathematics and Statistics

Abstract

This dissertation develops scalable Bayesian solutions for two critical challenges in modern data analysis: modeling failure times and imputing missing biological data. First, we introduce an adaptive semi-parametric MCMC framework for Weibull lifetime modeling, addressing the lack of conjugate priors and multidimensional sufficient statistics. Using hierarchical modeling and the No-U-Turn Sampler (NUTS) in STAN, we evaluate 24 prior combinations across 72 simulated datasets. The method yields robust parameter estimates under increasing and decreasing hazard rates and proves effective in predicting prostate cancer patient survival. Second, we assess imputation strategies for high-throughput proteomics data, where missingness distorts signal integrity. We compare MCMC, MICE, QRILC, and Random Forest methods using MAE, NRMSE, and correlation analyses. Among them, MCMC best preserves data structure across varying missingness and dimensionality. Together, these contributions demonstrate the versatility and robustness of Bayesian modeling for structured and unstructured data environments, offering practical tools for inference under uncertainty.

Share

COinS