Performance of single-trial classifications of viewed characters using EEG waveforms

Abstract

This paper examines the possibility of classifying characters viewed by subjects using single-trial Electroencephalogram (EEG) waveforms from the frontal and occipital areas of the brain. As a training data set, Event-Related Potentials (ERPs) were calculated for each character from the first 20 trials and the remainder were assigned to a test data set. To extract features of waveforms, the regression relationship between the EEG and ERP waveforms was calculated from the training data set using the Support Vector Regression (SVR) technique. As a measure of classification performance, cross-validation rates were calculated for the test data set and they incrementally increased with the number of channels when the regression relationship was used. This result provides evidence that this procedure using the relationship between EEGs and ERPs is effective in predicting viewed characters, and that performance can be improved by a combination of waveforms across electrodes.