Unsupervised discovery and validation of affective engagement states using synchronized EEG and eye-tracking during digital learning

This study explores affective engagement states in digital learning environments using synchronized EEG and eye-tracking data. EEG signals were collected from frontal channels and processed using sliding windows to extract band-power features. Frontal Alpha Asymmetry (FAA; log alpha power difference, F4–F3) and the Beta–Alpha ratio (BA; log beta/alpha power ratio) were used as affective proxies. Instead of predefined emotion labels, unsupervised clustering methods were applied to identify latent engagement patterns directly from EEG features. To ensure robustness, non-overlapping parity analysis and hold-out validation were performed. Statistical tests were conducted to compare FAA and BA values across the identified clusters. Results showed consistent differences in arousal-related features across clusters, supporting the reliability of the identified states. Eye-tracking data were synchronized with EEG windows to provide additional behavioral context. The findings demonstrate a robust multimodal framework for identifying and validating affective states during real classroom interactions.