Moment Analysis of Eeg Amplitude Histograms and Spectral Analysis: Relative Classification of Several Behavioral Tasks

Abstract

Previous studies indicate that EEG amplitude probability density functions are Gaussian (normal) during rest and non-Gaussian during performance of mental tasks. In the present study we compared measures of normality, including higher central moments (e.g., skewness, kurtosis) and relative spectral power, to classify data sampled from several different behavioral tasks (resting eyes closed and mental arithmetic). Analysis shows significant classification in 22 of 25 subjects, based upon a total of 46 EEG variables. However, only two of these variables involved Gaussian properties of the amplitude distribution. Relative spectral power, on the other hand, contributed 33 predictor variables in delta, theta, alpha, and beta frequency bands (alpha was the best single predictor). These results lend support to studies demonstrating the robustness of EEG relative spectra but cast doubt upon the utility of Gaussian patterns in EEG amplitude distributions as predictors of behavioral states.