Automatic Recognition of Human Psychological State Based on EEG Data

What is it about?

The paper deals with the problem of binary classification of Electroencephalography (EEG) data using ordinary personal computers at making computation in real time. Eleven different criteria of similarity of two Multivariate Time Series (MTS) were used for this purpose. Basis on computation results of 32 dimensional EEG signals was established the advantages of the considered methods over each other. Methods “ascending eigenvalue-weighted difference between eigenvector matrices”, “getting into the confidence regions of the linear trends of MTS” and of the method which is obtained by the union of previous two methods gave better results by classification accuracy than others.

Featured Image

Why is it important?

Binary classification of EEG data using ordinary personal computers at making computation in real time is considered. For this purpose here is considered eleven different criteria of similarity of MTS such as: Weighted Sum Singular Value Decomposition (WSSVD), Ascending Eigenvalue-Weighted Difference between Eigenvector Matrices (AEWDEM), Not Weighed Divergence Between Eigenvector Matrices (NWDEM), Not Weighed Divergence Between Covariance Matrices (NWDCM), Generalized Variance (GV), Descending Eigenvalue-Weighted Difference between Eigenvector Matrices (DEWDEM), Principal Components Analysis Similarity Factor (PCASF), Eros (Extended Frobenius norm), Dynamic Time Warping (DTW), Getting into the Confidence Regions of the Linear Trends of MTS (GCALT) and union of two AEWDEM and GCRLT methods. They were applied to the concrete data, in particular, to 32 dimensional EEG signals. Computing results show the superiority of AEWDEM, GCALT methods and the method, obtained by the union of these two methods in comparison with considered other methods.