Theses and Dissertations

Date of Award

8-1-2026

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Applied Statistics and Data Science

First Advisor

Kristina Vatcheva

Second Advisor

Santanu Chakraborty

Third Advisor

Zhuanzhuan Ma

Abstract

While substantial research has applied machine learning (ML) to cross-sectional data, comparatively little has evaluated these methods within longitudinal designs. This study compared traditional longitudinal statistical models with ML approaches for predicting type 2 diabetes (T2DM) using nurse-visit data from waves 2, 4, and 6 of the English Longitudinal Study of Ageing (ELSA) of adults aged 50 years and above. Generalized Linear Mixed Models (GLMM) and Generalized Estimating Equations (GEE) were compared with Random Forest (RF), Mixed-Effects Random Forest (MERF), and XGBoost under a common participant-level split, with thresholds optimized by cross-validation. RF, XGBoost, and GEE were closely matched on discrimination, as measured by AUROC (0.825, 0.825, 0.823), and exceeded GLMM (0.680). XGBoost achieved the highest PR-AUC (0.482) and best calibration; RF the highest sensitivity (0.631). Findings underscore the importance of threshold optimization and of matching modelling strategy to the temporal depth of repeated-measures data.

Comments

Copyright 2026 Peggy E. Akabuah. All Rights Reserved. https://proquest.com/docview/3409129285

Share

COinS