Immune clonal algorithm based feature selection for epileptic EEG signal classification

Yong Peng, Bao-Liang Lu

Abstract: Detecting epileptic EEG signal automatically and accurately is significant in evaluating patients with epilepsy. In this study, the immune clonal algorithm (ICA) is employed to perform automatic feature selection, reducing the number of features the classifier deals with and improving the classification accuracy. In the experiment, EEG signal was decomposed into five sub-band components by a discrete wavelet transform. Features were extracted as input to train three classifiers (NB, SVM, KNN and LDA) to judge whether the EEG signal was epileptic or not. Then, ICA was introduced to select a feature subset to train the classifiers. Experimental results show that the classification accuracy based on selected features is significantly higher than that on original features. We also analyzed the relative importance of each feature.