Back to search

Georgia Institute of Technology

Improvements in the Modeling of High Dimension/Low Sample Size Imbalanced Clinical Datasets

Abstract

dc:description.abstract

This dissertation has tried to tackle three common problems in the field of biomedical research data analytics. The problems that are covered in this document fall around the concept of imbalance and high dimensionality that is in the nature of the datasets that are gathered in this field of research, although they can be extended to other fields as well. Both imbalance and high dimensionality suffer this area of research by impacting the predictive models negatively resulting in the introduction of over-fit, or noise in the models. First high dimensionality of the whole blood gene expression arrays are addressed and a new approach using Stability Selection has been proposed to reduce the dimension of these datasets. Then a novel pipeline to combine single-cohort studies into multi-cohort studies is proposed. The pipeline is tested on two GSE datasets which verified the proposed approach. Then, Stability Selection was used to be combined with SMOGN to boost the performance of regression in imbalanced, small, and horizontal datasets. The increase in the regression accuracy enabled further discoveries on AirPICU dataset which showed the importance of PRISM as a very effective predictor of Ventilation Free Days which indirectly indicates the mortality chance. Finally, the use of unsupervised generative models like CTGAN and TVAE was investigated to reduce the imbalance in imbalanced datasets. It was observed that CTGAN is a powerful model that can improve the performance of SMOTE in imbalance removal both with over-sampling and complete synthetic data modeling. The new approach was tested on the cohort of ICU patients with a chance of experiencing fluid overload and it verified the positive impact of tabular GANs in imbalance correction of similar datasets.

Degree

thesis:*
Level thesis:degree_level
Doctoral
Department dc:contributor.department
Electrical and Computer Engineering
Grantor dc:publisher
Georgia Institute of Technology
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Ghiasi Rad, Milad
Advisors dc:contributor.advisor
  • Kamaleswaran, Rishikesan
  • Inan, Omer T.
Committee members dc:contributor.committeemember
  • Anderson, David V.
  • Grunwell, Jocelyn R.
  • Saedi, Soheil
  • Pan, Tony

Subjects

dc:subject × 4

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/1853/76785

Chain of custody

source
Harvested from
Georgia Tech
Base URL
repository.gatech.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Ghiasi Rad, Milad. Improvements in the Modeling of High Dimension/Low Sample Size Imbalanced Clinical Datasets. Doctoral thesis, Georgia Institute of Technology, 2023. https://hdl.handle.net/1853/76785