EEG Seizure Detection Using SVM Classifier and Genetic Algorithm

Abstract

Among all the neurological disorders, the epileptic seizure has become one of the most common disorders which affect people regardless of age. While electroencephalography (EEG) is the most common method to monitor the brain activity of a patient with epilepsy, an expert will be needed to analyze seizure activities. In this paper, a systematic method of seizure detection using classification and feature selection algorithms from EEG signals is proposed. The EEG signal from the database is processed through different stages by which features from the signal are extracted and selected. Genetic algorithm is very useful when there is large amount of data as it looks for solutions that otherwise couldn’t be found easily when compared to other feature selection methods. SVM is a simple and much efficient classifier algorithm with high efficiency and provides good classification performance than other classifiers. The selected features will be able to differentiate between seizure and non-seizure signals effectively with high accuracy. In the experimental results, KNN obtained the accuracy of 96.91% and sensitivity of 100% when no feature selection was used. By using ACO, accuracies of 92.59 and 91.88% are achieved by KNN and SVM, respectively. Among all the experimental results, the highest accuracy of 98.77% and sensitivity of 100% are achieved from the SVM classifier with GA as a feature selection method, whereas KNN achieved 94.44% accuracy.