DeepEnsemble: A Novel Brain Wave Classification in MI-BCI using Ensemble of Deep Learners

Arshad Mehtiyev, Aziz Al-Najjar, Hamidreza Sadreazami, Marzieh Amini

Abstract: Brain computer interfaces (BCIs) are rapidly growing in neurorehabilitation to realize communication for consumers with disabilities. This work proposes a classification approach for the motor imagery BCI electroencephalogram (EEG) signals using the recent advances in deep learning methods. The proposed DeepEnsemble method combines different deep learning models in an ensemble learning with soft voting. More specifically, multi-layer perceptron, vision transformer, convolutional neural network and its integration with distributed gradient boosting methods are explored separately and combined in an ensemble learning model. The proposed method is evaluated on a publicly available dataset. It is shown that the proposed ensemble learning method significantly improves the classification accuracy of the EEG signals across various subjects as compared to some of the existing methods.