A Multiclass Epilepsy Identification Technique Using Wavelet-Based Features

Epilepsy affects around 1% of the world's population. The electroencephalogram (EEG) is the most common measure of the brain's electrical activity. It is used clinically and by the research community to study brain disorders. This paper presents a comparative study of automatic detection of epilepsy using wavelet-based features with different classifiers. Both binary and multiclass classification setups are studied. The classifiers used are TreeBoost, multilayer perceptron (MLP) neural network, and support vector machine (SVM). Our study is evaluated using EEG dataset from the University of Bonn Hospital in Germany. The obtained results show the significance of different features block for the classifiers. In addition, the results show that TreeBoost outperforms other classifiers. In contrast to existing works carry only binary classification, we consider here 4 classes and show that our results are comparable to the results reported for the single 2-class problem.