Classification of brain signals based on proposed method

Abstract EN

Electroencephalography (EEG) signals are used to uncover brain processes, and researchers can use them to explore psychological factors that underpin behavior and perception.

Because EEG signals are aperiodic and non-stationary in nature, selecting features that are appropriate for a certain application is complicated.

Furthermore, EEG signals have a large number of dimensions, which decreases the speed and efficiency of the classifier.

The suggested approach for the categorization of brain signals aims to address such issues, and it varies from current approaches in that it skips the feature extraction .

To process the nature of non-stationary, aperiodic, or unstable EEG signals, a segmentation approach has been used.

Each one of the EEG channels is divided into N-segments, while each segment is divided into M-sub-segments.

The covariance matrix was utilized to minimize EEG's dimensions while maintaining the signal data's core features.

This is what allows the suggested approach to quickly become familiarized with all datasets.

A recognized classifier was employed to classify EEG data: Least Square Support Vector Machine (LS-SVM) .

The suggested system was put to test with the use of 4 datasets: a non-focal and focal epileptic dataset from the University of Bern-Barcelona, a motor imagery dataset (IVa) from brain-computer interface (BCI) competition III, a mental imagery tasks dataset (V) from BCI competition III, and epileptic dataset from Bonn University.

For each one of the datasets, LS-SVM classification results for sensitivity, accuracy, FPR and specificity are 100%, 100%, 0%, and 100%, respectively Keywords: Electroencephalography (EEG), Magnetoencephalography(MEG), Covariance- System., event-related brain potentials (ERPs), functional magnetic resonance imaging (fMRI) .