Classification of EEG Signals in Seizure Detection System Using Ellipse Area Features and Support Vector Machine

Epilepsy is a brain disorder, characterized by transitory and impulsive electrical signal of the brain. The electroencephalogram (EEG)-based seizure detection system used for automated diagnosis of epilepsy requires optimum classification of signals. This paper presents an optimized method for classification as normal and epileptic signals using Empirical-Mode Decomposition (EMD) technique. The Intrinsic-Mode Functions (IMFs) are few symmetric and band-limited signals obtained by applying EMD to the signals. However, optimum selection of IMF features is crucial step in deciding feature set for classification. The 95% confidence ellipse area is calculated from the Second-Order Difference Plot (SODP) of selected IMFs to form features for classification. The feature space is used by the Cosine Similarity Measure Support Vector Machine (CSM-SVM) classifier with optimum feature selection. It is observed that the features formed using ellipse area of dissimilar combination of IMFs have given superior classification performance on EEG data available from the Bonn University, Germany.