Spectral EEG features and tasks selection process: Some considerations toward BCI applications

In this paper, we further develop the idea of subject specific mental tasks selection process as a necessary prerequisite in any EEG-based brain computer interface (BCI) application. While, in two previous researches we proved - using the EEG-extracted auto-regressive (AR) parameters and twelve different mental tasks -, the major gains one can obtain in tasks classification performance only by selecting the proper tasks, here we investigate the putative relation that exists between each (subject, given EEG features) pair and the corresponding individual optimum set of cognitive tasks. In this idea, a set of three different spectrum relative power parameters were considered. The classification performances achieved with these last EEG features are comparatively presented for two subjects and for two sets of tasks: (i) the frequently used in the BCI field, Keirn and Aunon set of tasks, and (ii) the previously determined (AR-based) optimum individual set of tasks.