Epileptic Seizures Classification Using Burg Coefficients Based on Sparse Autoencoder and SVM Classifier

Abstract

Epileptic seizure classification with machine learning techniques has become the famous solution in the analysis and detection of epilepsy. Throw employing the smart algorithms on electroencephalogram (EEG) signals. Generally, these signals are non-stationary and complex. These techniques permitted to identify useful knowledge about epileptic seizures. Given this, the current paper introduces a novel methodology for automatic epileptic seizure classification and detection. In this approach, Burg method as an autoregressive model (AR) based on sparse autoencoder (SAE) with Support vector Machine (SVM) as a classifier have been used. First, features extractions are accomplish by Burg method. Then, the features given by such method are an input to (SAE). The fallouts of this later that employed in training and classification stage by using SVM classifier. To evaluate the performance of our framework, the Bonn database has been employed. In the experimental stage, three types of classification have been studied (healthy non-epilptic to Interctal or Ictal) and Interctal to Ictal. Also, two sorts of experimentations have been done. (i) the features extracted by Burg method then SVM classifier without SAE and (ii) the features selected by Burg method improved by SAE .The obtained results with an average accuracy for the first and the second experiments are (95.25 %,98.00%, 95.00%) and (94.50 %,100.00%, 96.00%) for the type of the classification studied respectively. They have clearly demonstrated the value of the introduced framework.