Identifying Emotional Signatures from Facial EMG Signals: A Clustering-Based Exploration (Preliminary Study)

Facial expressions are potent indicators of emotional states, yet interpreting electromyography (EMG) signals from facial muscles remains a complex challenge. This study investigates the potential of identifying emotional signatures from facial EMG signals through clustering analysis. Surface EMG data from seven facial muscles are analyzed in three versions: raw, filtered, and amplitude. Clustering algorithms, including K-means, Agglomerative Hierarchical Clustering, and DBSCAN, are applied to identify distinct patterns associated with different emotional states. Results indicate that K-means clustering on amplitude data yields the most promising patterns. Two primary clusters emerged: a majority cluster characterized by low amplitude distributions across all muscles and a smaller cluster exhibiting higher amplitudes, particularly in the zygomaticus and orbicularis muscles. These patterns do not align strictly with predefined video-based emotional categories (positive, negative, neutral), suggesting the intricate nature of facial expressions. The zygomaticus and orbicularis muscles demonstrated greater sensitivity to emotional changes, especially for positive stimuli. This research contributes to emotion recognition by establishing a foundation for more sophisticated emotion detection systems based on facial muscle activity. Further research is necessary to validate these findings on larger populations and with a wider range of emotional stimuli.