Machine Learning Methods for Decision Making Inference in Healthcare

Machine learning algorithms are widely regarded as disruptive innovations. They have demonstrated superior performance in many complex domains, such as computer visions, signal processing and natural language processing. One area, in particular, in which machine learning has potential widespread societal impacts is the healthcare delivery and policy making. Statistical and machine learning techniques, combined with big data generated from healthcare organizations, have the potential to bring changes to healthcare delivery. Despite major advances in the development of machine learning methodologies in the mainstream healthcare literature, several challenges remain. This thesis addresses two emerging challenges in the context of real-world healthcare applications, with a focus of developing novel machine learning methodologies. The first challenge is that big data does not guarantee reliable and valid results, without rigorous methodological support in data analysis. This thesis proposes novel and rigorous methods using large scale datasets in three application areas. In Chapter 2, I will present a framework that optimally extracts public online search information such as Google for accurate U.S. national and state-level COVID-19 and Influenza-like Illnesses (ILI) predictions. In Chapter 3, I will investigate two treatment-related decision-making problems using electronic healthcare records (EHRs) databases. In Chapter 3 section 1, I will introduce a unified propensity score method for causal inference analysis using EHRs. In Chapter 3 section 2, I will propose a reinforcement learning approach for optimal personalized treatment recommendations in ICU settings. The second challenge is that many traditional machine learning methods are not computationally efficient or feasible when applying them off-the-shelf in complex big data settings. In Chapter 4, I will present a computationally efficient parameter learning method in Hidden Markov Models (HMM), with application in sports-related concussion. In Chapter 5, I extend a previously developed dynamic systems inference method (Manifold Approximated Gaussian Process Inference) to situations with completely unknown systems dynamics, with applications in system biology and other scientific areas.

Level thesis:degree_level

Doctoral

Department dc:contributor.department

Industrial and Systems Engineering

Grantor dc:publisher

Georgia Institute of Technology

Year dc:date.issued

2023

Author dc:creator

• Ma, Simin

Advisors dc:contributor.advisor

• Serban, Nicoleta
• Yang, Shihao

Committee members dc:contributor.committeemember

• Xie, Yao
• Keskinocak, Pinar
• Ning, Shaoyang

• Machine Learning
• Deep Learning
• Healthcare
• Health Analytics
• Infectious Disease Prediction
• Online Search Behavior
• Spatial Temporal Modeling
• Causal Inference
• Propensity Score
• Electronic Healthcare Records
• Reinforcement Learning
• Precision Medicine
• Intensive Care Unit
• Concussion
• Hidden Markov Models
• Gaussian Process
• Ordinary Differential Equations

Language dc:language.iso

en_US

Handle dc:identifier.uri

https://hdl.handle.net/1853/71985

OAI identifier oai:identifier

oai:repository.gatech.edu:1853/71985