Recursive Least Squares based Seizure Detection of Epileptic EEG signals

Abstract

In this research paper, multiresolutional recursive least squares (RLS) method is proposed for the automatic epileptic seizure identification present in the electroencephalogram (EEG) signals and detection is performed using relative band power feature map. The main aim of this method is to improve the seizure classification accuracy of the epileptic EEG signals. In the proposed RLS based approach, initially the EEG signal is pre-processed using RLS dependent adaptive filter followed by the segment-wise computation of the relative band power in terms of frequency present in the epileptic seizure information. At last, support vector machine (SVM) is used to classify the EEG signal into seizure and non-seizure data. The experimental investigation shows that the RLS approach has attained a classification accuracy of 94.876%, 97.594% sensitivity, and 91.856% specificity. The proposed scheme has proven to be significant for EEG seizure detection, which will support doctors in the epileptic detection.