Approximation-Aided Epilepsy Detection Using Linear and Non-Linear Classifiers

Abstract

Patients with brain-related symptoms and diseases are diagnosed using electrocardiogram (EEG) signals. Epilepsy in humans can be diagnosed from EEG signals. This chapter focuses on identification of seizure-free, seizure, epileptic, and normal EEG signals with minimum-length EEG signal. The algorithm could classify the seizure and normal EEG signals even for a length of 1000 samples per segment. The algorithm was tested on various EEG signals. The traits are extracted from the EEG signal and preprocessed and fed to five different classifiers to check the accuracy of the scheme. The algorithm provided a better accuracy of 99.8945%. The sample signals were taken from an EEG signal database available at University of Bonn. The proposed scheme was tested with performance measures such as specificity (SPE), NPV (negative predictive value), PPV (positive predictive value), ACC (accuracy), MCC (Matthews's correlation coefficient), and sensitivity (SEN). The test results proved that the proposed methodology could perform real-time epileptic seizure detection.