NeuroPhysio: Explainable EEG-Based AI for Affective and Cognitive Load Recognition in Digital Health-Integrated Learning Systems

The ability of intelligent health-aware e-learning systems to recognize Affective States (AS) and Cognitive Load (CL) determines their success in personalizing instruction and improving learners' engagement. The proposedNeuroPhysioframework integrates an EEG-based analytic system that uses handcrafted feature extraction and deep neural model to study AS and CL for 3s and 5s windows of EEG. This research work applied statistical methods and entropy analysis to extract handcrafted features from frequency bands to identify brief and prolonged neurophysiological patterns during learning activities. The subject-dependent experimental results of 22 participants showed XGBoost achieved the high recognition accuracy of 88.00% for AS and 89.13% for CL assessment using a 5s window. The CNN model obtained good results by reaching 82.01% for AS and 84.83% for CL assessment of the same 5s window. The EEG-augmented models demonstrated exceptional accuracy in AS with XGBoost reaching 97.27% and for CL assessment CNN reaching 96.19%. The subject-independent results of the XGBoost and CNN models demonstrated stable performance within individual participants. However, their ability to generalize across participants showed a noticeable decline. TheNeuroPhysioframework for affect-awareness that relates EEG to an adaptive feedback system for health, technology, and education.