A Comparative Study of Epileptic Seizure Detection Framework using SVM and ELM

Abstract

Poverty and lack of health awareness are major reasons for illnesses, particularly neurology-related problems in India. Epilepsy is one such problem that affects the brain by causing seizures termed as epileptic seizures. People in rural areas believe epileptic attacks to be results of influence of black magic and resorted to unscientific practices for treatment. Repeated occurrence of seizures could lead to death. The early detection and treatment would cure 70 percent of the cases. The study of the epileptic activity can be done using EEG recordings of the brain. Although many software packages offers complete tool set for complex EEG analysis which is not as candid compared to brain-imaging techniques user need to choose appropriate framework suitable for their application scenario. In this paper we propose four different combination of feature extraction and classification techniques for detecting epileptic seizures and this study aims to compare the results in context of accuracy. The combination of discrete wavelet transform for feature extraction and early learning machine algorithm for classifications generates 90.1% accuracy in classifying epileptic seizures. Also this framework reduces computation time by selection of proper EEG channel data by preprocessing which helps to develop an expert system which emulates the decision making of a human expert.