Comparative Investigation of Different Classification Techniques for Epilepsy Detection Using EEG Signals

Abstract

Among the major brain abnormalities that have been identified, various remedial strategies are proposed to tackle most of such conditions. One of the serious abnormalities of the nervous system is epilepsy, which causes electrical distraction and strains the neural system. Usually, epilepsy is determined by the neurologist by analyzing the EEG signals grabbed from the brain. The task is very challenging as it requires continuous examination and connotation of the EEG signal of an epileptic patient. Hence, the development of efficient automatic systems is currently a bottom neck issue, which can recognize the epileptic seizure attack and distinguish between normal condition and epileptic seizure condition. Over the years, efforts have been made to establish an automatic system for precise detection and classification of an epileptic seizure. In the present work, a range of statistical, entropy, and fractal-based attributes are calculated from the coefficients of selected wavelet. The classifiers used in the paper are trained on the selected features as well as full features. The objective of the study is the selection of appropriate features and applying Tree, SVM, KNN, BPANN algorithm for the classification. Features are ranked with the help of Information Gain, Relief F, and Correlation attributes ranking techniques. Using Information Gain and Relief F feature ranking technique with 10-fold cross-validation, the classification accuracy of 100% was achieved by Backpropagation Artificial Neural Network (BPANN) for classifying epileptic and normal EEG signals.