Machine learning approaches in practical anxiety detection

This dissertation investigates the development and application of machine learning (ML) techniques for the robust detection of anxiety using wearable technology under diverse environmental conditions. The research encompasses three interlinked studies, each contributing uniquely towards advancing anxiety detection technologies in real-world settings. The first study conducts a thorough review of existing ML methodologies for anxiety detection, identifying the evolution from traditional feature-based (FB) models to advanced end-to-end (E2E) deep learning approaches. It evaluates their applicability across different scenarios, highlighting the challenges of integrating such technologies in practical applications due to issues like noise interference and model overfitting. In the second study, the focus shifts to the practical implementation of these models in noisy environments. It explores the resilience of both FB and E2E models by introducing artificial noise into the WESAD dataset and examining their performance. This part of the research emphasizes the critical impact of environmental disturbances on model accuracy, particularly in wearable technologies, and demonstrates the superior noise resistance of FB models compared to E2E models. The third study extends this analysis by applying transfer learning techniques to adapt these models to real-world datasets, specifically the RADWear and WEAR datasets, which reflect a broad spectrum of real-life conditions. The study assesses the efficacy of transfer learning in enhancing model robustness and addresses the challenges of deploying these technologies in dynamic and uncontrolled environments. Despite the high potential of transfer learning, the results reveal that E2E models consistently underperform in comparison to FB models, which display greater adaptability and reliability under varied environmental conditions. Overall, this dissertation highlights the complexities of developing effective ML-based anxiety detection systems that are capable of operating in real-world scenarios. It underscores the need for further research into optimizing model architectures, improving noise management strategies, and refining data collection techniques to enhance the practicality and effectiveness of anxiety detection tools. The insights gained from these studies pave the way for future advancements in wearable technology for mental health, offering a foundation for more personalized and responsive mental health care solutions.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Mechanical Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Alkurdi, Abdul E.

Contributors dc:contributor

• Hsiao-Wecksler, Elizabeth T
• Hernandez, Manuel E
• Shao, Chenhui
• Sowers , Richard

• Affective Computing
• Anxiety Detection
• Wearable Technology
• Ai For Health
• Applied Machine Learning

Statement dc:rights

• Copyright 2024 Abdulrahman Alkurdi

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125506

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125506