Plausibility assessment of a subject independent mental task-based BCI using electroencephalogram signals

In this research, we study the possibility of designing a mental-task based subject-independent Brain Computer Interface (BCI) using Electroencephalogram (EEG) signals. Due to major differences in the EEG signal of individuals during different mental tasks, designing a universal BCI seems impossible. Hence, almost all the previous studies concentrated on designing custom-based Brain Computer Interface systems (BCIs) which are appropriate to be used by only one particular subject. In order to overcome this limitation, this paper presents an efficient subject-independent procedure for EEG-based BCIs. The main aim of this research is to develop ready-to-use BCIs that can be applicable for all users. To achieve this goal, three feature extraction methods including Autoregressive modeling, Wavelet transform and Power spectral density were applied; then, a new method based on Genetic Algorithm (GA) wrapped Self Organization Map (SOM) feature selection was used to select the most related features with the use of leave-one-subject-out cross-validation strategy. According to the experimental results, the proposed algorithm based on GA wrapped SOM feature selection is an efficient method for designing subject-independent BCIs and is able to distinguished different cognitive tasks of different individuals, effectively.