Automatic classification of epileptic EEG signals based on AR model and relevance vector machine

Automatic classification system of epileptic EEG signals is one very important issue. In this paper a new epileptic EEG signal classification method was proposed on the basis of AR model and relevance vector machine. AR model was used to extract EEG features, and then principle components analysis and linear discriminant analysis were adopted to reduce the dimensionality of feature space. In order to obtain a sparser model and a model with probabilistic outputs, relevance vector machine was chosen as classifier. A publicly-available database was used to test the proposed method: the highest accuracy obtained in this paper is 99.875%; and even if the dimensionality of feature space is reduced to 1/15 of the original dimensionality, the classification accuracy was still able to reach 99.500%. The introduction of relevance vector machine makes the model sparser; the number of relevance vectors is just a few tenths of that of support vectors. The results mentioned above suggest that the method can be well applied in epileptic EEG signal classification.