Detection of epileptic seizures in eeg signals with rule-based interpretation by random forest approach

Abstract

Epilepsy is a common neurological disorder and characterized by recurrent seizures. Although many classification methods have been applied to classify EEG signals for detection of epilepsy, little attention is paid on accurate epileptic seizure detection methods with comprehensible and transparent interpretation. This study develops a detection framework and focuses on doing a comparative study by applying the four rule-based classifiers, i.e., the decision tree algorithm C4.5, the random forest algorithm (RF), the support vector machine (SVM) based decision tree algorithm (SVM + C4.5) and the SVM based RF algorithm (SVM + RF), to two-group and three-group classification and the most challenging five-group classification on epileptic seizures in EEG signals. The experimental results justify that in addition to high interpretability, RF has the competitive advantage for two-group and three-group classification with the average accuracy of 0.9896 and 0.9600. More importantly, its performance is highlighted in five-group classification with the highest average accuracy of 0.8260 in contrast to other three rule-based classifiers.