Mental task classification against the idle state: a preliminary investigation.

The motivation for this study was to obtain candidate electrode sites for use in online self-paced brain-computer interfaces and preliminary classification results for comparison to online tests. Six mental tasks were tested for classification against an idle state. Data representing the idle state was collected in association with active mental task data during each recording session. Features were extracted in two representations, band power and reflection coefficients. A sequential forward floating search algorithm was used to obtain prevailing electrode-feature pairs for each subject-task combination under two conditions: maximising classification accuracy and maximising mean trial accuracy. Methods used to select electrode-feature combinations are found to lead to differing electrode sites in a number of task-feature combinations. An across task prevalence towards electrodes positioned in the left frontal hemisphere is observed when maximising classification accuracy.