M.Sc. Tezi Görüntüleme

      Nowadays, the implementation of computerized systems in the medical field is quite popular. For this reason, the time taken for diagnosis of diseases and incorrect diagnoses have been decreased. These systems are also employed by the experts to improve treatment of illnesses. Because of these systems make storing and reinterpretation of diagnostic data more easier, the importance of development and support of these systems has increased.

      In this thesis, the data of healthy, pre-seizure and seizure period have been classified to diagnosis of epileptic seizure, and a comparative study of classifiers with classification accuracy have been made. For that, Hjorths Descriptors, Recurrence Quantification Analysis, Shannon Entropy, Hurst Exponent, Zero Cross Rate methods, which have been used previously for diagnosis of disease or analysis of EEG signals, have been utilized. Obtained attributes have been firstly classified using k-nearest neighbor, naive Bayes, neural networks and support vector machines severally. For further performance improvement, the methods are evaluated in hybrid manner. For both detecting epileptic seizure, and distinguishing healthy and diseased people, 7 sub-data sets are created and accuracy, sensitivity and specificity of the methods were determined for the data sets. In this case, classification accuracy of between 97.5% and 100 % has been obtained and the results compared with the literature show that significant contributions have been made with this study.

      Keywords: Epileptic Seizure, Hjorth, Entropy, Recurrence Plot, Hurst Exponent, Hybrid Classifier