Towards a Cross-Participant Cognitive Load Classification Using Eye Tracking and Deep Learning

Cognitive Load (CL) is a critical cognitive construct in many sectors and fields, such as cognitive science and humancomputer interaction (HCI). Yet achieving reliable real-time measurement of CL remains challenging. Eye tracking has been shown as a noninvasive and deployable physiological signal for inferring CL, but few studies show generalized CL classification performance using eye-tracking alone and there is limited understanding of which eye-tracking features should be used. Therefore, this study assessed the viability of raw pupillometry and gaze features for generalized CL prediction using a machine learning approach. Eye-tracking data was collected at 60Hz from 89 participants performing the N-back task, with was binarized into low (0–1 back) and high (2–3 back) CL conditions. Performance was assessed using inter-subject testing. Results show that both XGBoost and a modified Vision-Transformer showed performance exceeding 75% indicating cross-participant generalizability, with the Vision-Transformer reaching 85% when combining pupil and gaze features. These findings support the feasibility of using eye-tracking and machine learning for generalizable CL estimation. Future studies should examine generalizability under varying ambient conditions and in real-world tasks.