EEG based epilepsy detection using approximation entropy and different classification strategies

Epilepsy is the 4
^th
prevalent neurological disorder, which distinguishes itself by frequent seizures. These seizures are unprovoked and can be detected by investigating the electroencephalogram (EEG) of the subject. In this paper, an automatic method is proposed to discriminate between normal and epileptic (interictal& ictal) classes. Approximation entropy (APEN) has been utilized as a single informative feature. Multiple classifiers such as: support vector machine (SVM), Probabilistic Neural Network (PNN) and Soft-max regression along with different strategies for classification have been examined. Experiments were carried out by the most popular dataset used in this domain. 96.875%, 53.75% and 65% are the best results achieved by soft-max regression classifier using multi-class strategy for normal, interictal and ictal classes respectively.