Dynamic Mode Decomposition Approach for Binary Motor Imagery EEG Signal Classification

In this work, a synthetic minority oversampling technique (SMOTE) is employed in electroencephalogram (EEG) signals for Data-Augmentation (DA). Then, the Dynamic Mode Decomposition (DMD) method is leveraged to separate the augmented EEG signals into amplitude, frequency, and modes for Motor Imagery (MI) EEG signal classification. The features are acquired from augmented data and applied to a fine decision tree with 5-fold cross-validation for classification. For a fair comparison of the proposed algorithm, publicly available dataset 2b of Brain Computer Interface (BCI) competition IV is used. To verify the effectiveness of the DA technique, the features are fed to the classifier with and without the DA technique. The proposed method achieves enhanced classification accuracy for binary MI EEG signals contrasted with existing pioneering approaches.