Beyond averaging II: Single-trial classification of exogenous event-related potentials using stepwise discriminant analysis

Using stepwise discriminant analysis (SWDA), single-trial event-related potentials (ERPs) were classified as to whether they were elicited by a checkerboard presented to the upper or lower visual half-field. Discriminant functions were computed on the basis of ‘training sets’ constructed of upper and lower half-field ERPs, and applied to ‘test sets’ of other ERPs elicited by the same stimuli. Individual-subject discriminant functions for data recorded at Pz classified the single ERPs in the test sets with a mean accuracy of 83.7% correct. The mean accuracy attained by individual-subject functions from the most discriminable scalp site for each subject was 87.8% correct, and that attained by an across-subjects function was 78.1% correct. Averaged ERPs showed the previously reported polarity reversal of corresponding exogenous components in the upper and lower half-field wave forms. Moreover, the SWDA procedure chose ERP time points at the latencies of these exogenous components for discriminating the half-field ERPs. The results demonstrate that SWDA can accurately classify single ERPs in which the systematic variance is localized in exogenous components, having periods within the range of frequencies which typically compose the background EEG