Date of Award

Spring 2023

Project Type


Program or Major

Electrical and Computer Engineering

Degree Name

Doctor of Philosophy

First Advisor

Md Shaad Mahmud

Second Advisor

John LaCourse

Third Advisor

Edward Song


The medical field produces large quantities of multidimensional, complex, and often unstructured data. The timely interpretation of medical data is crucial for medical diagnostics. Machine learning applications are increasingly becoming popular with the ever-increasing computational ability and data availability. Machine learning methods can potentially find complicated relationships that may not be apparent in traditional clinical decision rule systems used in medical applications. Models deployed in medical applications need to be reliable, time efficient, and easy to interpret, thus making working with biosignals challenging in some cases.

The first part of the dissertation focuses on building predictive systems using existing data repositories. It is shown that deep learning models have the potential to improve the predictive capability of using multidimensional data. Next, we developed our system to predict the Math Anxiety Score of College Students using Functional Near Infra-Red Spectroscopy (fNIRS) and Physiological Signals. The experiment setup, signal processing, and data modeling for this application is presented in this dissertation. One of the reasons that clinical decision rules are widely used is due to their simplicity and being easily explainable. We will discuss how machine learning models can be interpreted and which modalities and features have high importance when determining Math Anxiety Levels. For some applications collecting data can be both times consuming and expensive task. In such cases, data augmentation methods used in machine learning can generate synthetic data samples. Finally, data augmentation methods for both classification and regression models are discussed, along with how these methods improve the predictive capability of the models with examples using existing data and our data.