Enhanced Subspace Alignment with Clustering and Weighting for Cross-Subject Multi-Session EEG-based Emotion Recognition

The non-stationary nature of brain activity signals and their many inter-subject differences have created many challenges in the practical applications of electroencephalogram (EEG)-based emotion recognition, such as brain-computer interfaces. In such a way, using traditional classifiers in classifying these signals would significantly decrease accuracy when applying the classifier to a new subject. Domain Adaptation methods seem to be an effective way to solve this problem by minimizing the difference between the EEG signals of different subjects. But in the basic techniques for domain adaptation, looking at all subjects’ data in the same look causes the loss of a part of the potential power of these methods. The present study proposed an extended version of the Subspace Alignment method for Domain Adaptation utilizing the clustering of source subjects. This method can make an effective choice between the available data from the source domain by clustering the source subjects based on the behavioral similarity of their EEG signals. This effective selection among the subjects and the idea of purposeful data weighting have caused the final model to represent the space of the target subject more effectively. To confirm the feasibility of the proposed method, we have conducted experiments on the SEED dataset, which is widely used in the emotion recognition EEG-based task, and it achieved an accuracy of 84.3%. The results show that the presented model has better classification accuracy than several state-of-the-art models.