A comparative study of machine learning algorithms for epileptic seizure classification on EEG signals

Electroencephalography (EEG) 1s a tool for monitoring brain activity which is important for identifying epilepsy seizure. Automatic epileptic seizure identification in EEG is a challenging task and useful for helping neurophysiologists. This study compares some algorithms in machine learning algorithm that combine features extraction and classification algorithm for epilepsy seizure identification based on EEG data. The classification algorithms compared in this study are Generalized Relevance Learning Vector Quantization (GRLVQ), Backpropagation, SVM, and Random Forest, combined with Wavelet and PCA feature extraction. The EEG signals used in this study were obtained from EEG dataset which was developed by University of Bonn. EEG epilepsy seizure dataset has five classes. Class A and B are from five healthy subjects in open and closed eyes. Class C, D, and E from five elliptic subjects, where C and D are no-seizure signals, and E contains only seizure signal. The tasks that are used to compare the performance of feature extraction and classification algorithm is classifying 5 classes of EEG epilepsy seizure on EEG dataset. The measurements for evaluating methods are: accuracy, recall, precision training and testing times. The best feature extraction method at our experiment is PCA. The best performance in recognizing the five classes in EEG epileptic seizure dataset is GRLVQ, with the accuracy, precision and recall is 0.9866 and testing time is less than 0.1 seconds.