Classification of P300 in EEG signals for disable subjects using singular spectrum analysis

Abstract

Brain-computer interfaces have been enabled severely disabled users to communicate with their environments. One method is to use a controlled stimulus to elicit the P300 event-related potential. EEG signals during the repeated stimuli were recorded from four disabled subjects and processed with a Butterworth bandpass filter and Singular Spectrum Analysis, normalized, separated into 2 groups of the target and non-target trial data, and averaged for every 5 trials for each group before classified using a neural network. The purpose of averaging every five target and non-target trials was to emerge the P300 component of even-related potentials so that the target trials could be differentiated from the non-target trials. Further processing by selecting 1 of every 5 processed non-target trials increased the value of sensitivity by 10.9%, it showed that the number of false negatives of target trials was reduced. The results of the classification gave the maximum accuracy of 92.5%. The average values of sensitivity, specificity, and accuracy were 70.8%, 89,8%, and 84.6% respectively.