Application of Multi-domain Fusion Methods for Detecting Epilepsy from Electroencephalogram Using Classification Methods

Abstract

Electroencephalogram (EEG) signal is a time series delineative signal which contains the useful knowledge about the state of the brain. It has high temporal resolution for detection of chronic brain disorders such as epilepsy/seizure, dementia, etc. Technically, a feature mainly targets to capture the significant and typical characteristics hidden in EEG signals. In view of the low accuracy of commonly used methods for discrimination of EEG signals, this paper presents an efficient multi-domain fusion method to enhance classification performance of EEG signals. Features are extracted using autoregressive method (AR) employing Yule-Walker and Burg’s algorithms respectively to generate feature from EEG. This paper implements two schemes of multi-domain fusion methods, the first one is AR method and wavelet packet decomposition (WPD) and the second one is AR method and Sample entropy (SampEn). Next, classification of extracted features is performed by different classifiers like Support vector machine (SVM) classifier, Linear Discriminant Analysis (LDA) classifier, Artificial neural network (ANN) classifier, K-nearest neighbor (KNN) and Ensemble classifier. Compared to AR-based method, fusion methods are yielding high accuracies. The ANN classifier has obtained the highest classification accuracy of 98.12% with the feature AR Burg-WPD combination compared to other classifiers in multi-domain fusion methods.