In order to improve accuracy of mental task classification

we propose a new method of EEG classification with feature extraction.First

the raw signals are decomposed by wavelet packet decomposition (WPD).Then

using wavelet packet entropy reflecting the distribution of signal energy in time and frequency domains

the best basis of wavelet packets is selected from a wavelet packet library according to the wavelet packet entropy.Afterwards the statistical features are used to represent the best basis wavelet coefficients.Moreover

the eigenvector is obtained by calculating the asymmetry ratio of the hemispheric brainwave at each electrode in different mental tasks.Finally

the performance of the eigenvector is evaluated via a support vector machines classifier.A publicly available EEG database was used to validate this study.Compared to the conventional WPD

wavelet packet best basis decomposition and existing autoregressive feature extraction methods

the average accuracy for the proposed method ranged from 95.41% to 99.65% for ten different combinations of five mental tasks.