Study on the effect of different frequency bands of EEG signals on mental tasks classification

Currently, frequency bands not more than 40 Hz are usually used to perform mental tasks classification in brain-computer interface systems. In this study, by using Keirn's EEG data, we studied the effects of ten 10 Hz-wide subbands between 0 and 100 Hz on mental tasks classification. Features were computed in frequency domain as the sum of weighted power spectral value in each subband at each channel (C3, C4, P3, P4, O1, and O2). Fisher's linear discriminant was used to perform task-pair classification. Our results indicated that subbands ranging from 30 to 100 Hz resulted in relatively greater classification accuracy at many scalp sites. The average classification accuracy of 98.3% across 130 task pairs was achieved by using features including those obtained on gamma bands (30-100 Hz), which is much greater than that of 89.3% by using the frequency band 0-30 Hz only.