Entropy Analysis as an Electroencephalogram Feature Extraction Method

Abstract

The aim of this study was to evaluate a possibility for using an entropy analysis as an electroencephalogram (EEG) feature extraction method in brain-computer interfaces (BCI). The first section of the article describes the proposed algorithm based on the characteristic features calculation using the Shannon entropy analysis. The second section discusses issues of the classifier development for the EEG records. We use a support vector machine (SVM) as a classifier. The third section describes the test data. Further, we estimate an efficiency of the considered feature extraction method to compare it with a number of other methods. These methods include: evaluation of signal variance; estimation of spectral power density (PSD); estimation of autoregression model parameters; signal analysis using the continuous wavelet transform; construction of common spatial pattern (CSP) filter. As a measure of efficiency we use the probability value of correctly recognized types of imagery movements. At the last stage we evaluate the impact of EEG signal preprocessing methods on the final classification accuracy. Finally, it concludes that the entropy analysis has good prospects in BCI applications.

Keywords

• brain-computer interfaces (BCI)
• electroencephalogram (EEG) analysis
• entropy analysis
• Shannon entropy nu- support vector machines (nu-SVM)