Pupil Diameter Classification using Machine Learning During Human-Computer Interaction

The significance of eye tracking and pupillary responses spans various disciplines, especially for people with severe disabilities. Effortful decision-making is marked by pupil dilation, reflecting increased cognitive load and can be used as a potential measure for system adaptation to users' mental states. This study investigates the classification of pupil diameter data to differentiate between decision-making and focus time in a mobile robot navigation task. Data were collected from 19 healthy participants utilizing an eye-tracking-based user interface to control the robot's movements along pre-set paths. This paper presents a deep learning and SVM-based classification approach to distinguish focus and decision-making from pupil diameter patterns, offering insights for achieving an adaptive command selection approach. On average, the proposed deep learning model has an average accuracy of over 82% in classifying the data for participants using pupil diameter data.