Theses and Dissertations

Advisor

Wang, Haifeng

Committee Member

Smith, Brian

Committee Member

Gonzales Vargas, Jessica

Committee Member

Ball, John E.

Committee Member

Chander, Harish

Date of Degree

12-12-2025

Original embargo terms

Immediate Worldwide Access

Document Type

Dissertation - Open Access

Major

Industrial and Systems Engineering

Degree Name

Doctor of Philosophy (Ph.D.)

College

James Worth Bagley College of Engineering

Department

Department of Industrial and Systems Engineering

Abstract

Parkinson's disease (PD) is a complex condition with a wide range of clinical symptoms. It is a progressive neurological disorder that has afflicted an estimated 1 million people in the US and 10 million worldwide. The diagnosis of PD is typically based on the presence of clinical features, with no specific diagnostic test or biomarker. The methods of assessment for PD are also used, in whole or in part, for similar symptom diseases such as Multiple Sclerosis, Essential Tremors, Multiple System Atrophy, Supranuclear Palsy, Dementia with Lewy bodies and Huntington’s disease. Many of the current clinical tests have low sensitivity and specificity, leading to misdiagnosis in as many as one in four cases. Machine learning has stepped into the gap of diagnostic tools for PD, Neural Networks being a popular option. In this research, a LeCun-Particle Swarm Optimization hybrid weight initialization method is combined with gait parameters to train a Neural Network that improves model accuracy over existing methods by up to 3%. Gait feature bilaterality is shown to be an important model design consideration, as well as the need to pair the type of model algorithm with the type of data available. The improvements achieved are expected to have implications on future gait feature selection and machine learning modelling efforts.

Download

Share

COinS