Automatic Detection and Classification of Epileptic Seizure using Radial basis Function and Power Spectral Density

Abstract

An attempt made to detect and classify the epileptic seizure using a simple procedure is suggested. The radial basis function is applied to the Electro Encephalo Gram (EEG) data and statistical features such as mean, variance, standard deviation and root mean square are extracted. The mean of the power spectral density is computed from the EEG data. The statistical features along with mean of the power spectral density are fed into the classifiers such as Support Vector Machine classifier (SVM), K-Nearest Neighbour classifier (KNN) and Naive Bayes classifier (NB). The classification process includes K-fold cross validation. The dataset used is publicly available in Bonn University and it contains 5 subsets which include S, Z, O, N and F. The subsets include EEG recordings from seizure affected patients and normal persons. The SVM and KNN classifiers perform better for classification of subsets containing normal EEG recordings and seizure recordings producing the accuracies of 98.93% and 99.05% for two fold cross validation and 98.99% and 99.30% accuracies for four fold cross validations respectively. For the classification of seizure and seizure free recordings obtained from hippocampal formation in opposite hemisphere in brain, the SVM classifier outperforms the other two classifiers, producing the accuracy of 97.16% and 97.27% for two fold and four fold cross validation respectively. In the classification of seizure recordings and seizure free recordings obtained from epileptogenic zone, NB classifier performs well with accuracies of 91.16% and 90.39% for two and four fold validations respectively.