EEG signal classification for epilepsy diagnosis based on AR model and RVM

In this article, we propose a new EEG signal classification method based on Relevance Vector Machine (RVM) and AR model. It can well separate the ictal EEG signals from the inter-ictal ones, this is very important in the diagnosis of epilepsy. Our studies can be divided into three parts: firstly, EEG features were extracted from the signals based on AR models, and then the performance of these features was evaluated; secondly, according to the performance of the features, feature selection was introduced between feature extraction and classifiers; finally, RVM was implemented with different AR models, different kernel widths, and different subsets of the features in order to get an overview of the method. The results indicate that: (1) features extracted based on AR models can well represent the EEG signals in the task of EEG signal classification for epilepsy diagnosis; (2) feature selection is needed between feature extraction and classifiers; (3) the method based on RVM and AR model can well differentiate the two types of EEG signals.