Date of Award
9-22-2025
Document Type
Thesis
Degree Name
Mathematics, MS
First Advisor
HaoYang Teng
Committee Members
Hong Zhou; Mohamed Milad
Abstract
This thesis examines change-point detection in statistical models through the framework of two-way truncated linear regression with threshold effects. Our work is based on Two-Way Truncated Linear Regression Models with Extremely Thresholding Penalization (TWT-LR with ETP) by Teng and Zhang (2022) which is an innovative approach extends traditional linear regression by incorporating potential change-points in both intercept and slope parameters, effectively handling multiple change-points through a specialized penalization technique that shrinks small coefficients to zero. This methodology demonstrates remarkable effectiveness in identifying significant change-points while maintaining computational efficiency. We propose a framework that models both upper and lower thresholds in predictor variables directly. Our methodology is applied to CDC’s COVID-19 mortality data. Using carefully selected parameter ranges and increment sizes, we balance accuracy and computational efficiency in threshold estimation. Results reveal distinct threshold effects in predictors such as age and total death counts, providing valuable insights into COVID-19 mortality risk factors.
Rights Management
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License
Recommended Citation
Vu, Hoang, "Applied Linear Regression Techniques to Biomedical Research and Detecting Change-Points to Enhance Model Flexibility by Improving Interpretability and Applicability" (2025). Student Theses and Dissertations. 1089.
https://arch.astate.edu/all-etd/1089
