Classification of EEG signals using fuzzy classifier for brain-computer interface

Abstract

The paper presents a fuzzy system for the classification of electroencephalogram (EEG) signals developed for an asynchronous brain-computer interface (BCI). The main purpose of the work is to study the applicability of fuzzy logic systems for EEG classification aimed at recognizing both physical movements and imaginary motor commands. The proposed approach to EEG signal classification is based on the feature extraction method that calculates mutual information between EEG channels after signal preprocessing. This method is aimed at identifying significant correlations between signals obtained from different brain regions, allowing forming of feature vectors which reflect the relations between EEG channels. A fuzzy classifier was built for 72 % of the subjects for physical movements and 48 % of the subjects for imaginary motor commands. During the experiments, it was found that the average accuracy is about 74 % for physical movements and 60 % for movements performed mentally. The best result for an individual reaches 96 % and 71 % accuracy for physical movements and movements performed mentally, respectively. The influence of preprocessing parameters on the system efficiency was predicted using classification trees. It was found that the most significant predictors are the decimation frequency during preprocessing of EEG signals and the number of bins during the calculation of mutual information between EEG channels for both real movements and imaginary motor commands. The accuracy of the forecast reached 73 % and 86 % for physical movements and movements performed mentally. Thus, the proposed method is interesting from the point of view of the analysis of EEG signals and feature generation. The work demonstrates the possibilities of fuzzy logic for creating brain-computer interfaces, but highlights the need for further research in the field of optimizing preprocessing parameters and modification of the linguistic scales of the fuzzy classifier to improve both the accuracy and interpretability of the system.