Using Attentive Network Layers for Identifying Relevant EEG channels for Subject-Independent Emotion Recognition...

Emotion recognition approaches using electroencephalographic (EEG) signals can be evaluated using two distinct approaches: subject-dependent and subject-independent. While subject-independent models are more generalizable and practical than subject-dependent models, they face challenges due to the high variability of EEG signals among individuals. One solution is identifying shared patterns during emotional processing. As deep learning has become a common practice for emotion recognition, using attentive network layers can help identify shared predictive patterns. This study explores this approach by using attentive network layers to identify brain areas relevant for predicting four emotions evoked by video clips in 15 individuals. The model achieved an average accuracy of 46% (95% CI: 41.3-50.7%) among subjects, indicating that the EEG channels in the right hemisphere were more relevant for predicting happy and neutral emotions, while those in the left hemisphere were more relevant for sadness and fear. These findings highlight the importance of including EEG channels from both hemispheres to ensure the prediction of different emotion types in subject-independent approaches. Clinical relevance- This study identifies shared neuronal patterns in emotion prediction, supporting the development of generalizable emotion recognition systems that can help diagnose and treat disorders such as depression, anxiety, and neurodegenerative diseases.