Combination of frequency bands in EEG for feature reduction in mental task classification

Brain-computer interfaces require online processing of electroencephalogram (EEG) measurements. Therefore, speed of signal processing is of great importance in BCI systems. We present a method of feature reduction by combining frequency band powers of EEG, in order to speed up processing and meanwhile avoid classifier overfitting. As a result a linear combination of power spectrum of EEG frequency bands (alpha, beta, gamma, delta & theta) was found that reduces the dimension of feature vector by a factor of 5. This method gives a total correct classification rate of 91.71% comparing to 87.96% achieved from direct use of frequency band powers and 85.54% achieved from PCA feature reduction method applied to the same feature vector with 14 components.