Fast Seizure Detection from EEG Using Machine Learning

Abstract

A seizure is a sudden, uncontrolled electrical disturbance in the brain. It can cause changes in epileptic patient's behavior, developments or emotions, and in levels of consciousness. The rapid predictions of epileptic seizures help the epileptic patient to avoid tremendously complications like falling, drowning, accidents and pregnancy complications. Generally, seizure detection is performed in two main sequential stages; feature extraction stage and classification stage. In this paper, new algorithm is proposed to detect the seizure through a short period only 10 seconds. Eleven features are extracted from EEG signal to characterize behavior of EEG activities. These features are fed to five classifiers. These classifiers are SVM, KNN, decision tree, logistic regression, and ensemble. The results show that SVM is the best classifier for detecting the incidence of the seizure with high accuracy and sensitivity.