Title: Automated EEG-based epilepsy detection using BA_SVM classifiers

Authors: Aya Naser; Manal Tantawi; Howida A. Shedeed; M.F. Tolba

Addresses: Faculty of Computer and Information Science, Ain Shames University, Cairo, Egypt ' Faculty of Computer and Information Science, Ain Shames University, Cairo, Egypt ' Faculty of Computer and Information Science, Ain Shames University, Cairo, Egypt ' Faculty of Computer and Information Science, Ain Shames University, Cairo, Egypt

Abstract: Epilepsy is a neurological disorder which affects individuals all around the world. The presence of epilepsy is recognised by seizures attacks. EEG signals can provide useful information about epileptic seizures. Unlike most of the existing studies which consider only two classes, this paper proposes an automatic EEG-based method for epilepsy detection which has the ability to distinguish between the three classes; normal, interictal (out of seizure time) and ictal (during seizure). In the proposed method, Rènyi entropy, line length and energy are computed from each of the five sub-bands extracted from an EEG segment using digital wavelet transform (DWT). Thereafter, the extracted features are fed into BA-SVM classifiers trained using divide and conquer strategy for classification. The BA-SVM classifier is a support vector machine (SVM) classifier whose parameters are optimised using BAT algorithm. The popular Andrzejak database was utilised for training and testing purposes. The average accuracies for all considered cases are more than or equal 95%. Thus, the various experiments and comparisons accomplished in this study reveal the efficacy of the proposed method.

Keywords: electroencephalogram; EEG; epilepsy; digital wavelet transform; DWT; entropies; support vector machine; SVM; bat optimisation.

DOI: 10.1504/IJMEI.2020.10032889

International Journal of Medical Engineering and Informatics, 2020 Vol.12 No.6, pp.620 - 625

Received: 26 Jul 2018
Accepted: 29 Jan 2019

Published online: 06 Nov 2020 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article