EEG signal analysis for Seizure detection using Discrete Wavelet Transform and Random Forest

Epilepsy, a recurring disorder is symptomized by unprovoked seizures. Considered as one of the most common neurological disorders, Epilepsy affects people of all ages. Around 65 Million people across the world suffer from this disease. Manual diagnosis of EEG signals of long duration may be a source of error as well as a cumbersome task. Hence automation in Seizure Detection is essential for diagnosis of Epilepsy. Therefore, in this paper, an expert system for EEG signal classification has been proposed. The proposed system aims at classifying EEG signals into 3 classes- Normal, inter-ictal (EEG recordings of epileptic patients during non-seizure period) and ictal (EEG recordings of epileptic patients during seizure period). Discrete Wavelet Transform, a technique for analyzing signals in time-frequency domain has been quite successful in this respect for extraction of features from EEG signals. Further Envelope Analysis has been used on the DWT coefficients before feature extraction in order to elevate the performance of the system. Finally several Machine Learning techniques (E.g: MLP and Random Forest) have been used for classification of the EEG signals and the relative accuracy of these Classifiers with regard to this problem have been compared. The results from the experiments show that the Random Forest Classifier is the most effective classifier for this problem delivering an accuracy of 98%. The data that has been used for this work has been taken from the publicly available EEG database of University of Bonn which presents 100 single channel EEG records for each of the above classes.