Comparative Study on Machine Learning Classifiers for Epileptic Seizure Detection in Reference to EEG Signals

Abstract

Epilepsy is the key concern of medical practitioners and machine learning researchers since last decade. EEG signals play a very crucial role in early detection of epilepsy as well as cure of epilepsy. The traditional approach to analyze EEG signals includes two main steps: feature extraction and classification. Since multi-channel EEG data is chaotic data, selecting optimal features and classifying them are major challenges. There exist a number of feature extraction and classification techniques proposed by researchers which perform well. For feature extraction, wavelets have been proved to perform state-of-the-art performance, but no such state-of-the art performance exists for classification techniques. The classifiers explored in the presented work include random forest classifier, support vector machine, Naïve Bayes, k-nearest neighbor, decision trees, artificial neural network, and logistic regression. Experimental results on the UCI dataset represent that random forest is performing best with 99.78% accuracy.