General Framework for Multi-Classification of EEG Signals Based on Multi-Scale Properties

Abstract

Numerous computer automated diagnosis (CAD) systems have been proposed to detect epilepsy in electroencephalogram (EEG) signals. The aim of this paper is to look at multi-scaling properties obtained by multi-scale analysis (MSA) as main distinctive features to simultaneously distinguish between all categories of EEG signals that compose the popular database hosted by the department of epileptology, University of Bonn, Germany. Particularly, multi-scale analysis is employed to capture long-range properties of the EEG signal at different scales used to represent its short and long variations. Then, the obtained multi-scale properties are used to train four different classifiers; namely, k-nearest neighbor (k-NN), linear discriminant analysis (LDA), naïve Bayes (NB), and the support vector machine (SVM). Experimental results based on ten-fold cross-validation method show that each single classifier achieves 100% accuracy. In this respect, multi-scale properties are found to be effective as they outperformed existing works on the same database by achieving perfect accuracy to distinguish between all five distinct EEG categories. Overall, the obtained results are promising.