Total Variation Based Multi Feature Model for Epilepsy Detection Using Support Vector Machine

Abstract

ABSTRACTEpilepsy is a critical brain disease which occurs primarily based on changes or connectivity issues between two neurons. Electroencephalography (EEG) is an efficient modality for capturing brain signals with respect to stimuli. Total variation (TV) is an energy based feature which helps to discriminate epilepsy from normal pattern. This paper proposes an efficient epileptic detection system based on TV model. Existing methods use single feature for classification, whereas, combination of multiple relevant features helps for better understanding of epilepsy. This paper also discusses combined multi feature model with TV for efficient detection and classification of EEG signal subbands (gamma, beta, alpha, theta, and delta) using support vector machine. Results show 100% delta band based classification accuracy for combination of features such as TV and sample entropy. This research also shows that combination of multi features such as, Shannon, spectral, TV, and linear predictor coefficient gives good classification accuracy for most of the subbands.