Unsupervised Clustering of Individuals Sharing Selective Attentional Focus Using Physiological Synchrony.

Area of Science:

• Neuroscience
• Cognitive Science
• Signal Processing

Background:

• Monitoring attentional states in real-world settings is challenging due to limited labeled data.
• Traditional methods require specific stimuli knowledge and labeled data for training attention models.

Purpose of the Study:

• To investigate unsupervised clustering combined with physiological synchrony for automatic identification of shared attentional focus.
• To assess the efficacy of electroencephalogram (EEG), electrodermal activity (EDA), and heart rate in distinguishing attentional groups without prior knowledge.

Main Methods:

• Applied dimensionality reduction and unsupervised clustering to physiological synchrony data (EEG, EDA, heart rate).
• Analyzed unimodal (EEG, EDA, heart rate separately) and multimodal (combinations) data.
• Used data from 26 participants listening to an audiobook with interspersed sounds, with differing attentional instructions.

Main Results:

• Unsupervised clustering on EEG physiological synchrony achieved a maximum accuracy of 85% in identifying attentional groups.
• Multimodal approaches combining EEG with EDA and/or heart rate generally yielded higher accuracies than EEG alone.
• Multimodal data classification was more consistent across different algorithms compared to unimodal data.

Conclusions:

• Unsupervised classification of attentional groups based on physiological synchrony is feasible, even without knowledge of stimuli or explicit labels.
• Combining EEG with other physiological measures enhances the accuracy and robustness of attention monitoring in naturalistic settings.
• This approach supports future research on attentional engagement in everyday environments.