Error Reduction Methodology and Data Simulation for Interval Data

Author

• Ranik Christopher Jelinek, Utah State University

Date of Award

12-2025

Degree Type

Thesis

Degree Name

Departmental Honors

Department

Mathematics and Statistics

Abstract

Chronic kidney disease (CKD) is a progressive condition affecting hundreds of millions of individuals worldwide. However, clinical datasets often record continuous laboratory measurements as categorical intervals rather than precise numerical values. This interval-censored structure presents methodological challenges for standard regression-based classifiers. This study compares three strategies for handling interval-valued predictors prior to fitting a logistic LASSO model: (1) midpoint imputation, which replaces each interval with its arithmetic center; (2) ordinal encoding, which maps intervals to integer ranks; and (3) a Monte Carlo simulation approach, which repeatedly samples uniformly from each observed interval and averages predictions across replications. Using a 10-fold cross-validated LASSO with area under the ROC curve (AUC) as the tuning criterion, all three methods achieve near-perfect classification on the CKD dataset (AUC > 0.997). DeLong’s tests indicate no statistically significant difference in discriminative performance between methods. These results suggest that, in high-signal clinical settings, the choice of intervalimputation strategy may have limited impact on predictive performance. Nevertheless, the Monte Carlo approach provides conceptual advantages in uncertainty quantification and may offer greater utility in noisier or lower-signal datasets.

Recommended Citation

Jelinek, Ranik Christopher, "Error Reduction Methodology and Data Simulation for Interval Data" (2025). Undergraduate Honors Capstone Projects. 1122.
https://digitalcommons.usu.edu/honors/1122