Expectation-Maximization Extreme Machine Learning Classifier for Epileptic Seizure Detection

The electroencephalogram (EEG) signal contains important physiological characteristics that are used to study the human brain operation. It is used to identify epileptic seizures and diagnose the epilepsy. The physical analysis of huge volume of EEG signal by a human specialist take more time and often create discrepancies in decision making process. In this work Expectation Maximization Extreme Learning Machine (EM-ELM) based technique is adapted for perfect detection of epileptic seizure in EEG signals. The suggested strategy utilizes discrete wavelet transform for subband decomposition of EEG signals. In the consecutive stage, attributes such as line length, tonal power ratio, and weighted 1D-Local Binary Pattern (LBP) has been extracted in the decomposed EEG signal. In the final stage EM-ELM classifier has been trained to classify seizure and conventional EEG signal. The performance of the system is compared with Bayesian, Gaussian Mixture Model and Decision Tree classifier. The maximum accuracy of the proposed method is 95.5%.