Comparison of Linear and Quadratic Classification of Event-Related Potentials on the Basis of Their Exogenous or Endogenous Components

The linear and quadratic techniques for classifying exogenous and endogenous components of event-related potentials are compared. The test and train on the same data, test and train on different data, and the leave-one-out techniques of evaluating classifier performance are employed. Time samples from waveforms measured from four electrodes simultaneously were used to design the classifiers. Classification results utilizing both techniques gave comparable accuracies suggesting that differences in the data covariances were not significant for the test conditions occurring in each paradigm.