Classification of P300 component in single trial event related potentials
H.O. Gulcar, Y.K. Yilmaz, Tamer Demıralp · 2002
In order to classify the P300 wave in single trials of an auditory oddball paradigm, an artificial neural network based on backpropagation error learning algorithm is implemented. After training, the neural network is expected to classify the responses into two categories according to the applied rare (target) and common (non-target) stimuli types. To prevent overfitting, early stopping and 10-fold cross-validation are applied. A simple data purification method, then, is suggested and applied to purify the data set before training the neural network. After purification, the neural network shows an improved performance of 96% correct classifications.