Next-Generation Healthcare with Applied Machine Learning and Quantum Machine Learning
Date Issued
May 2026
Author(s)
Onim, Md Saif Hassan
The University of Tennessee
Advisor(s)
Himanshu Thapliyal
Additional Advisor(s)
Travis Humble
Rebekah Herrman
Jinyuan Sun
Abstract
Older adults often require non-intrusive monitoring tools, as their physiological and emotional responses are influenced by age-related changes, cognitive vulnerability, and psychological and social factors. Traditional approaches such as self-reports or facial recognition can be intrusive, subjective, or difficult to use for individuals with cognitive impairment. This dissertation proposes a computational framework for identifying digital biomarkers of stress, emotion, and related psychosocial states using physiological signals from wearable devices, integrating classical machine learning, deep learning, and hybrid quantum machine learning methods. The framework begins with context-aware stress detection using both classical machine learning and deep learning approaches. Here, feature-engineered supervised learning techniques combine electrodermal activity, blood volume pulse, heart rate variability, acceleration, and skin temperature with task-level contextual information to establish a practical classification performance. Expanding from detection, time-series prediction using recurrent neural networks and long short-term memory models demonstrates the feasibility of stress forecasting and reveals temporal patterns that could support proactive interventions. Lastly, emotion-intensity estimation models based on regression techniques enable the independent modeling of emotional responses that can reduce individual variability and lay a foundation for reliable digital biomarkers. To address the challenge of modeling complex, high-dimensional physiological relationships, the framework incorporates quantum machine learning techniques. Physiological signals are embedded into quantum feature spaces to improve class separability and robustness to noise. Hybrid quantum models further support anomaly detection by identifying potential tampering or corruption in physiological data streams. Preliminary results indicate certain trials in these models outperform classical baselines even with modest sample sizes, showing promise for real-world healthcare and cyber-physical monitoring. The dissertation also investigates the scalability of quantum models by examining performance as qubit counts increase and feature encodings extend to higher-dimensional representations. Systematic evaluation under Noisy Intermediate-Scale Quantum constraints analyzes trainability, generalization, and computational cost. This work establishes practical guidelines for scalable quantum-enhanced digital biomarker modeling. Together, these contributions provide a foundation for personalized, continuous physiological monitoring in older adults and at the same time, advance theoretical and empirical understanding of hybrid quantum-classical machine learning in healthcare applications.
Degree
Doctor of Philosophy
Major
Computer Engineering
File(s)![Thumbnail Image]()
Name
ONIM-PRIMARY-2026.pdf
Size
13.52 MB
Format
Adobe PDF
Checksum (MD5)
74c38d03ee830c3d256bb411d343fae5