Improving Multi-Class Motor Imagery EEG Classification Using Overlapping Sliding Window and Deep Learning Model

Motor imagery (MI) electroencephalography (EEG) signals are widely used in BCI systems. MI tasks performed by imagining doing a specific task and classifying through EEG signal processing. However, it is challenging to classify accurately. In this study, we propose LSTM-based classification framework enhance accuracy of four-class signals. To obtain time-varying data signals, sliding window technique used, an overlapping-band-based FBCSP applied extract the subject-specific spatial features. Experimental results on competition IV dataset 2a showed average 97% kappa value 0.95 all subjects. It demonstrated that proposed method outperforms existing algorithms for EEG, also illustrates robustness variability inter-trial inter-session data. Furthermore, extended experimental channel selection best performance when using twenty-two channels method, but 0.93 was achieved with only seven channels.