Feature Vectors Combination to Increase the Efficiency of a Brain-Computer Interface (BCI) System

Brain-computer interface (BCI) system as a modern way of communication between brain and external device has got much attention in the recent decades. The basis of BCI is to record brain signals using one of the invasive (such as ECoG) or non-invasive (such as EEG) methods,to interpret different states of the brain and to make control orders for external devices.Preprocessing, feature extraction, feature reduction and data classification are the stages of processingbrain signals which are of great importance in a BCI system. In this paper, by implementing the stages of processing brain signals, the type of movement imagery is determined and using the method of features combination and superior features selection, the error rate of predicting the type of movement is optimized. It must be mentioned that in this paper, the dataset number IIB of the brain-computer competitions (2008), recorded from nine persons,is used. After preprocessing and eliminating signal artifacts, feature vectors are extracted by the parametric model of AR, AAR, wavelet transform and CSP filter. The main goal of this paper is to investigate the efficiency offeature vectors separately and to provide a method for BCI system efficiency increase; to achieve this goal, first the feature vector effects for each person are investigated using SVM classifiers; then by combining feature vectors the BCI system is tested and to increase its efficiency the methods of superior features selection are used. Investigating tables and figures we will conclude that by combining feature vectors, brain signals classification precision related to movement imagery may be increased.