Determination of epilepsy from different time scale EEG signs

Epilepsy is a common disorder in approximately 1% of the world population, which adversely affects the quality of life of people and repeats itself with continuous bouts. This disorder is a serious condition that occurs in some or all of the human brain, sudden, unexpected and irregular electrical discharge server. Electroencephalography (EEG), one of the instruments used in the detection and diagnosis of epilepsy, is an important source for information about epilepsy. EEG is widely used because of its speed, cost, accessibility and ease of use. Because of these features, EEG has become a research topic that many researchers are interested in. Many pattern recognition systems have been developed for automatic detection of epilepsy with EEG signs. In this study, an automated pattern recognition system is presented using A and E sign sets obtained from healthy and epileptic seizure subjects taken from Bonn University database. The pretreatment system presented in this study, has three stages, namely feature extraction and classification occurs. The aim of this study is to realize the different time-scale detection of EEG signals from epilepsy. For this purpose, in the first stage, 4096 samples and 23.6 seconds A and E sign sets, 128, 256, 512, 1024, 2048, 4096 samples and corresponding 0.7375, 1.475, 2.95, in order to obtain different time-scale EEG signals, 5.9 divided into 11.8, 23.6 second segments. A(128), A(256), A(512), A(1024), A(2048), A(4096), E(128), E(256), E(512), E(1024), E(2048) and E(4096) were obtained. In the second stage, spectral information of EEG signals have been reached by using Peridogram and Welch Power Spectral Density (PSD) estimation methods. In this study, different spectral methods and behavior patterns of EEG signaling segments were tried to be investigated. After the PSD estimation, the average of all the EEG signals was applied to reduce the data size of the marks obtained and each segment was represented by the feature vector at 16 sample lengths. The third step obtained for each EEG segments and K-using the feature vectors of 5-fold cross-validation method in 16 sample length Nearest Neighbor Algorithm (K-NN), Support Vector Machine (SVM) and Excessive Learning Machine (ELM) are classified. Performance achievements for all classifiers support the literature. Performance of different classifiers, the EEG signal segments of different feature vectors obtained by using a plurality of classifiers with different PSD estimates study were studied. Consequently, the results are examined and some suggestions in terms of work that can be done to shed light on the future.