Classification of Epileptic EEG Signals using Wavelet-EMD-Domain Features and Improved Multi-class SVM 1

Epilepsy is a chronic neurological disease caused by a disturbance in the electrical activity of the brain. In this paper, a novel approach based on multi-domain features, a selection of significantly important features from three domains namely time, wavelet and wavelet-EMD and combining them to classify EEG signals for epileptic seizure detection is proposed. The statistical time-domain features such as minimum, maximum, mean, standard deviation from time-domain, non-linear features namely largest Lyapunov exponent (LLE), approximate entropy (AE), and correlation dimension (CD) from the wavelet-EMD domain are extracted and used for the classification process. For feature selection, inter-class and intra-class based entropy is applied. Appropriate class-specific features that characterize the sub-band are selected to improve classification accuracy and to reduce computation time. An improved multi-class support vector machine is employed for the classification of epileptic EEG signals. The performances of the proposed methods are evaluated using two different benchmark EEG datasets such as Freiburg and Bonn. The performance measures namely classification accuracy, sensitivity, specificity, execution time and receiver-operating characteristics (ROC) are used to evaluate and analyze the performances of the proposed classifier. It is learned from the experiments conducted that the proposed method provides better performance in terms of improved classification accuracy with reduced execution time compared to that of the existing methods.