2025 Theses Doctoral

Multimodal neurophysiological measurement of cognitive load during math problem solving in 4th graders

Zeng, Jialing

Educational technology environments increasingly emphasize personalized and adaptive learning, yet accurately identifying learners’ moment-to-moment cognitive states to optimize instructional strategies remains challenging. Cognitive load, the amount of load imposed on an individual’s cognitive system when performing a task, has been recognized as a focus in understanding learner’s cognitive states and guiding instructional designs within educational technology environments. The advancement of non-invasive physiological sensors combined with artificial intelligence techniques shows the potential to enable the recognition of cognitive load dynamics through streaming biometric data collection and analysis, without disrupting learning experiences. Despite these promising technological developments, current literature remains limited in ecological validity, providing relatively few insights applicable to meaningful educational contexts.

By leveraging insights from brain science to education, this dissertation study investigated two primary objectives towards cognitive load state recognition: (1) exploring neurocognitive correlates of cognitive load using EEG measures, recording brain electrical potentials, and eye-tracking measures, recording ocular responses; and (2) developing and evaluating a multimodal physiological signal-based framework for recognition of cognitive load states in educational contexts. A two-phase within-subjects experimental design was employed, involving synchronous EEG and eye-tracking recordings as children performed math arithmetic tasks varying in complexity. These tasks were designed to induce two cognitive load states. Experiment 1 (82 fourth graders) explored underlying neurocognitive mechanisms of cognitive load variations and validated physiological indicators. Experiment 2 (37 fourth graders) collected a separate set of EEG and eye-tracking data five months later under comparable conditions, validating the cognitive load recognition framework informed by Experiment 1.

Results indicated significant neurophysiological correlates of cognitive load. Specifically, increased task complexity induced higher cognitive load, accompanied by reduced alpha power, increased theta power and theta/alpha ratios, and elevated sample entropy. Eye-tracking metrics showed increased pupil diameter and decreased gaze transition entropy under higher complexity conditions. Convergent validity analyses demonstrated small to moderate yet significant correlations among physiological, subjective, and performance-based cognitive load measures, supporting EEG and eye-tracking as objective indicators of cognitive load. Machine learning models using selected multimodal neurophysiological signals exhibited reasonable predictive capabilities, with Logistic Regression and Support Vector Machine classifiers achieving accuracies of 77.21% and 75%, respectively. These findings underscore the efficacy of multimodal physiological signals for cognitive load state recognition, contributing substantively to cognitive load theory and informing practical applications in adaptive educational technology.