Date of Award
2026
Degree Type
Dissertation
Degree Name
Doctor of Philosophy in Electrical Engineering
Department
Electrical, Computer, and Biomedical Engineering
First Advisor
Kunal Mankodiya
Second Advisor
Dhaval Solanki
Abstract
Chronic neurodevelopmental and neurological conditions are typically managed through episodic in-clinic evaluations that cannot capture the continuous symptom fluctuations needed to optimize treatment and medication. Wearable devices enable continuous at-home monitoring of psycho-physiological signals such as ECG and accelerometry, along with derived measures like step count, physical activity, sleep, and stress. Clinical translation, however, remains constrained by the lack of three things: deployable machine learning (ML) infrastructure for real-time inference on resource-constrained hardware, behavioral monitoring algorithms validated in free-living settings, and publicly available, context-aware, annotated datasets from consumer-grade devices. This dissertation addresses these gaps through three healthcare applications that test the feasibility of ML-integrated wearable systems. This dissertation addresses the following research question: How can machine learning be integrated into wearables to monitor behavioral markers of health applications using quantitative psychophysiological parameters?
Section 1 addresses the challenges of motor symptom tracking in unconstrained environments. To facilitate in-home assessments, this research introduces a customized smart glove ML-based system. Leveraging adaptive signal processing, the system extracts critical motor features, such as movement speed and amplitude, to quantify the impacts of medication intake on Parkinson’s Disease symptoms. To process this data efficiently, an Edge-Fog-supported telehealth IoT infrastructure was developed. By deploying ML classifiers directly onto the fog infrastructure, this section demonstrates how localized computing can support low-latency telehealth applications for continuous movement assessment and symptom management.
Section 2 addresses at-home cognitive and behavioral monitoring through the design of a digital health platform. This section introduces MindGame, an Internet of Medical Things (IoMT) puzzle game that leverages computer mouse interaction dynamics and commercial smartwatch data to identify behavioral patterns associated with Attention Deficit Hyperactivity Disorder (ADHD). Through at-home feasibility studies, specialized data pipelines were developed to analyze mouse cursor trajectories. This section provides a rigorous evaluation of wearable data reliability, establishing parameters for generating high-quality monitoring data outside of laboratory settings. This section also introduces a stimulated in-lab study (FidgetSense) to identify hyperactive behavior patterns in children with ADHD (N=20 children).
Section 3 focuses on the challenges of acquiring, visualizing, and annotating the massive datasets generated by multimodal wearables. This section first presents a custom data acquisition architecture combining a textile-based chest belt and a smartwatch system validated through an extensive 31-participant study designed to capture physical and mental stress markers. To bridge the gap between datasets and interactive annotation, this work concludes with the development of BioViz Studio. This containerized architecture, BioViz Studio, provides a portable environment for the rendering and annotation of multi-million-sample physiological time-series datasets across diverse sensor modalities.
Collectively, this dissertation advances ML-integrated wearable systems along three complementary dimensions: efficient on-device and edge-fog inference for motor symptom tracking, validated behavioral monitoring in free-living settings, and the datasets and open-source tooling needed to make such work reproducible. Altogether, these contributions move consumer-grade wearables closer to clinically actionable, continuous monitoring, and the publicly released datasets, applications, and annotation framework provide a foundation for future research in wearable digital health.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Recommended Citation
Sadhu, Shejar, "APPLIED MACHINE LEARNING FOR WEARABLE DIGITAL HEALTH TO MONITOR HUMAN BEHAVIOR AND PSYCHO-PHYSIOLOGICAL PARAMETERS" (2026). Open Access Dissertations. Paper 4601.
https://digitalcommons.uri.edu/oa_diss/4601
Included in
Biomedical Engineering and Bioengineering Commons, Electrical and Computer Engineering Commons