A Remark on the Most Informative EEG Signal Components in a Super-scalable Method for Functional State Classification based on the Wavelet Decomposition

Abstract

Technologies for automated real-time classification of functional states are essential for supervision over operators of critical infrastructure, stress resistance evaluation, functional studies of sportsmen. They also open new horizons for automated educational applications and applications for phobia therapy, especially supported by virtual reality technologies, as they allow the automated adaptation of tasks in real time. In 2012 a novel approach for the functional state automated classification based on electroencephalographic (EEG) data was introduced by E.D. Livshitz et al. The approach efficiently utilizes CDF 9/7 wavelet decomposition instead of classical Fourier analysis and provides a promising classification reliability. In this paper connections between a set of estimators that were identified as the most informative for this approach and expert knowledge used in EEG data analysis and manual functional state classification are studied. The formalization of these connections is based on a fact that in spite of substantial differences between CDF wavelets and standard trigonometric functions used in Fourier analysis wavelet-based estimators have a good localization in frequency domain.