Optimizing EEG Signal Classification with Neural Network Models

The study aims to improve motor imagery classification using EEG signals by leveraging sophisticated neural network architectures, including TransformerNet, LSTM-GRU, BiLSTM-GRU, Residual- GRU, and BiRNN-LSTM. The study focuses on the onset of four motor imagery movements involving the left fist, right fist, both fist, or both feet. The comparative analysis across these models revealed that the BiLSTM-GRU architecture achieved superior performance, with an accuracy of 71.67%. The BiLSTM-GRU model exhibited exceptional capability in discriminating between diverse motor imagery states, reflected in its leading accuracy and other performance metrics. The findings underscore the effectiveness of the BiLSTM-GRU architecture in improving the accuracy of cognitive state assessment and augmenting the performance of brain-computer interface (BCI) systems, representing a considerable advancement in the domain of neural signal processing.