Enhanced electroencephalogram signal classification: A hybrid convolutional neural network with attention-based feature selection

The enhanced EEG model achieves an impressive average classification accuracy of 85.53% for each subject using motor imagery electroencephalogram signals.

• Traditional machine learning methods struggle with motor imagery electroencephalogram signals due to inherent challenges like nonlinearity and low signal-to-noise ratios.
• An automatic feature extraction method using deep learning was developed to improve MI-EEG classification.
• Noise reduction techniques, including discrete wavelet transform and common average reference, were applied to the original MI-EEG signals.
• A convolutional neural network was utilized to extract time-domain features, while spatial features were also extracted to understand brain activity relationships.
• The proposed model improved classification accuracy compared to existing methods, with increases of up to 11.24% over CNN and 11.18% over CNN-LSTM.

AI simplified