Back to search

University of North Texas

From Missing to Meaningful: Advancing Imputation Methodologies Using Single and Multi-Model Machine Learning Techniques

Abstract

dc:description

In the healthcare system and the cochlear implant industry, one of the significant challenges is the inconsistency caused by the use of different test metrics across clinics and insurance companies. For instance, while some audiology clinics prefer the Consonant Nucleus Consonant (CNC) test, many insurance providers favor the Arizona Bioindustry Association (AzBio) test. These discrepancies lead to data inconsistencies that often result in additional costs for patients and insurers, while the greatest concern remains the lack of a fast and accurate method for converting results between tests. This study addresses these challenges by developing a predictive model that establishes correlations between various speech recognition tests (AzBio and Bamford-Kowal-Bench (BKB)) and word recognition tests (CNC), providing a practical and reliable solution that can be easily deployed to streamline clinical workflows. Another critical challenge in hearing healthcare is the presence of missing data in audiograms, which hampers accurate diagnosis and treatment. To improve the accuracy of audiogram imputations, this study implements several machine learning techniques, including Linear Regression, Lasso, Linear Support Vector Regression, Bayesian Ridge, K-Nearest Neighbors (KNN), and Random Forest. These models were used to impute missing audiogram frequencies ranging from 125Hz to 8000Hz, using data from over 4000 audiograms of 800 unique patients, including cochlear implant recipients and candidates. Results from Nested Cross-Validation demonstrate that Random Forest models achieved the highest accuracy, with Root Mean Square Error values between 4.74 and 6.37 and R2 values ranging from 0.91 to 0.96. Similarly, KNN models achieved RMSE values between 5.00 and 7.72 and R2 values between 0.89 and 0.95. Additionally, predictive classification models trained on datasets imputed with the best-performing models outperformed those using constant imputations by approximately 2%. To further enhance imputation accuracy, this research introduces Methodology of Multi-Model Imputation (MMMI), a novel approach that combines KNN and Random Forest models to improve the reliability and consistency of audiogram data imputation. By leveraging KNN’s ability to capture local relationships and Random Forest’s strength in handling complex interactions, MMMI achieves superior performance compared to traditional imputation techniques. Comparative analyses reveal that MMMI delivers a 2-20% improvement in prediction accuracy over individual models, demonstrating its potential for broader applications in medical data imputation. The results underscore the effectiveness of these models in ensuring more accurate, complete, and reliable datasets, ultimately improving patient outcomes and decision-making processes in audiology and beyond.

Degree

thesis:*
Grantor dc:publisher
University of North Texas
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Beaver, Sarah Anne
Contributors dc:contributor
  • Boyce, LeAnn
  • Morozov, Kirill
  • Bryce, Renee C.
  • Bozdag, Serdar

Subjects

dc:subject × 5

Rights

dc:rights
Statement dc:rights
  • Public
  • Beaver, Sarah Anne
  • Copyright
  • Copyright is held by the author, unless otherwise noted. All rights Reserved.
Language dc:language
English

Identifiers

dc:identifier.*
Identifier
local-cont-no: submission_4390
https://digital.library.unt.edu/ark:/67531/metadc2481315/
ark: ark:/67531/metadc2481315

Chain of custody

source
Harvested from
University of North Texas
Base URL
digital.library.unt.edu/oai/
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Beaver, Sarah Anne. From Missing to Meaningful: Advancing Imputation Methodologies Using Single and Multi-Model Machine Learning Techniques. University of North Texas, 2025. https://doi.org/10.12794/metadc2481315