Classification of EEG signals represented by AR models for cognitive tasks - a neural network based method

In this paper, the discrimination of mental tasks by means of the EEG signals is transformed into classification of a system that has as the output the EEG signals. A feedforward neural network is trained to classify six-channel EEG data into one of five classes which correspond to the selected tasks. A simpler topology of the neural network and a reduction of the dimension of layers are achieved due to an autoregressive (AR) model used to represent EEG signals. The network performances were analyzed based on classification rate for the cross-validation set.