Brain-computer interfaces (BCIs) employ EEG signals to offer a promising pathway for individuals with movement impairments to regain communication and control abilities. However, subject-independent motor imagery EEG classification remains a challenge due to differences in EEG signal distributions across subjects. This leads to reduced classification accuracy for new subjects. As a solution to this challenge, we propose a two-stage approach. In the first stage, we present ensemble representation learning (ERL) method to project EEG data into a new feature space. This ERL method aims to learn representations from EEG signals that are less prone to inter-subject variability and better suited for classification across different subjects, thereby improving classification accuracy. In the second stage, we utilize a hybrid classification model that combines the strengths of k-nearest neighbor (KNN) and support vector machine (SVM) algorithms to accurately classify the learned representations. Experiments on BCI competition IV Datasets 1 and 2a show that the proposed approach achieves average accuracies of 85.21% and 85.65% for new subjects, respectively, which outperforms the previous state-of-the-art methods. The results reveal that the effectiveness of our approach for subject-independent EEG classification in brain-computer interfaces.
Hamidreza Hosseinzadeh (Tue,) studied this question.