A General Purpose Approach to BCI Feature Computation Based on a Genetic Algorithm: Preliminary Results

BCI systems aim at developing man-machine communication channels independent of the intervention of muscles. This is accomplished by recognizing specific mental states and using their detection to trigger actions in a computer controlled environment. Brain activity is acquired, typically through EEG, and is then processed in order to compute features allowing the classification of the user’s mental states being monitored. Several successful approaches to BCI based on different neural mechanism underlying the generation of the signal patterns to be recognized can be found in the literature. Yet, the signal processing leading to such meaningful features may be quite diverse between the different approaches due to the variability introduced by the different subjects, acquisition devices, experimental setup. Here we developed a new, general purpose approach to the computation of features allowing efficient trial classification based on a genetic algorithm. The algorithm was tested on three different datasets drawn from the BCI competition II and based on slow cortical potentials, motor imagery and self-paced movements.