EEG Signal Classifiaction with Optimized LSTM: A Meta-Heuristic Approach

Abstract

How to exploit the potential relationships between EEG signal dimensions is one of the major focus in brain-computer interface (BCI) research. This study proposes a unique electroencephalographic (EEG) classification model with four processing steps (a) preprocessing, (b) feature extraction, (c) optimal feature selection, and (d) EEG signal classification Preprocessing of the obtained raw data is performed using Wiener filtering. Feature extraction is performed on the preprocessed data. For the most important features, such as Blackman window short-time Fourier transform (BF-STFT), discrete cosine transform (DCT) and discrete wavelet transform (DWT) are extracted from the preprocessed data. Combining these features, the most optimal features are selected using a custom spiral update whale optimization algorithm (CSUWOA). The optimized features are used to train a deep learning classifier. The optimized LSTM represents the different classification steps of the EEG data. Finally, the classification results of the EEG signals are the result of long short-term memory (LSTM). We used a custom spiral update whale optimization algorithm (CSUWOA) to fine-tune the weights of the LSTM, aiming to improve the detection accuracy of the model. Finally, we performed a comparative study to confirm the efficiency of the model.