Discriminating mental tasks using EEG represented by AR models

Abstract
EEG signals are modeled using single-channeland multi-channel autoregressive (AR) techniques. The coefficientsof these models are used to classify EEG data intoone of two classes corresponding to the mental task the subjectsare performing. A neural network is trained to performthe classification. When applying a trained network totest data, we find that the multivariate AR representationperforms slightly better, resulting in an average classificationaccuracy of about 91%.I....