Year: 2026
Type: dissertation
Abstract: The ability to express emotions to others is an important component of human social interaction and overall well-being. However, individuals with severe motor and communication impairments (SMCI) often face barriers when conveying their emotions to unfamiliar communication partners. The overall aim of this thesis is to move the needle forward in the development of emotion recognition technologies for individuals with complex communication needs. A systematic review and meta-analysis were undertaken to identify current research in emotion differentiation via unimodal eye-gaze. Findings indicate that recurrent use of shared databases is hindering progression in terms of generalizability of results and advancement of machine learning approaches. In addition, the importance of reporting metrics such as precision, recall and F1 score for machine learning models was identified. Next, metrics of eye-gaze were studied via classic statistical analyses and filter methods for feature selection prior to machine learning. Results showed the importance of pupillometry features such as maximum pupil diameter and saccade velocities in emotion differentiation. A study with 28 typically developing (TD) individuals was completed to identify the efficacy of unimodal eye-gaze technologies in emotion differentiation. Results indicated that while an increase in the number of emotion categories decreased accuracy and F1 scores, more categories may be able to better capture the nuances of human emotion. The next study focused on the clinical population of interest, with recruitment of 18 individuals with SMCI and age/sex matched TD individuals to better explore targeted training and testing of machine learning algorithms. The accuracy of machine learning models improved when individuals with SMCI were included in the algorithm training process. Based on the challenges faced throughout the recruitment and data collection process of data from persons with SMCI, ten recommendations are provided for future research. The contents of this thesis will serve to inform future studies on emotion recognition via physiological signals and assist researchers facing similar recruitment and data collection challenges.
Author: Collins, Mackenzie Lenora
Institution: Queen's University
Language: English
Topic: Emotion and Mood Recognition
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Solve ME/CFS Initiative
OpenAlex ID: W7207971583
