Optimisation of Real-time EEG Signal Classification Accuracy

Abstract

This study proposes a hybrid model based on the combination of a convolutional neural network (CNN) and a long short-term memory network (LSTM) for the classification of electroencephalographic (EEG) signals, with a particular focus on the motor imagery task in the BCI Competition IV Dataset 2a. The usability of the data is enhanced by band-pass filtering and independent component analysis (ICA), which effectively removes artefacts and noise from the signal, thus improving the quality of the data. In the process of feature extraction, time-domain statistical features are employed, while high-dimensional features are reduced in dimension through principal component analysis (PCA), thus enhancing the computational efficiency and classification performance of the model. This paper experimental results show that the CNN-LSTM model achieves 100% accuracy on the training set, but only 72.41% on the test set. These findings suggest that the model demonstrates robust classification capabilities when processing training data, but exhibits some limitations in its ability to generalise to previously unseen data. This paper also discusses the limitations of the model and makes suggestions for improvements to further optimise its generalisation performance and enhance the results of its applications.