Multimodal Deep Learning for Cognitive Fatigue Detection in E-Learning Using Eye-Tracking and EEG
Abstract
Multimodal Deep Learning for Cognitive Fatigue Detection in E-Learning Using Eye-Tracking and EEG
Incoming article date: 12.12.2025Recent growth in online learning has created a need for reliable methods to monitor learner engagement, cognitive load, and fatigue. This study presents a deep learning framework that integrates eye-tracking data with electroencephalogram features to classify engagement levels in digital learning environments. Eye-tracking indicators of cognitive load, including pupil dilation, blink rate, fixation duration, and saccade velocity, were extracted from a publicly available dataset and combined with electroencephalography (EEG) measures. Engagement level was modelled as a three-class problem, including low, moderate, and high, using hybrid CNN-LSTM architecture designed to capture both spatial and temporal patterns. The model achieved an overall accuracy of approximately 89 percent with high precision and recall across categories. ANOVA analysis showed that no single feature could reliably distinguish engagement levels, underscoring the benefit of multimodal deep learning. The study highlights how combining eye-tracking measures with EEG signals can offer a clearer, real-time picture of learners’ cognitive states during e-learning activities. By detecting moments when attention declines or cognitive fatigue begins to set in, such systems can enable genuinely adaptive learning platforms, ones that know when to suggest brief breaks, adjust the pace of instruction, or provide timely, targeted support to help learners stay engaged.
Keywords: cognitive fatigue, deep learning, e-learning, eye-tracking, student engagement, EEG