The Automatic Detection of Epileptic Seizures Based on EEG Signals Processing: Investigation of Different Features and Classification Algorithms

Abstract

Automatic detection of epileptic seizures has been extensively studied and documented in literature. However, the topic continues to be of interest as reliable algorithms for general use are still being investigated. The challenge comes from the complex nature of the EEG signal and of the epileptic seizure, as both show patient specific characteristics. This makes highly performing algorithms developed on specific datasets difficult to translate to a more general use case. To provide more insights into the characteristics of seizure and non-seizure EEG segments, this paper proposes and investigates several features. Feature combinations are selected and fed per patient to both an Support-Vector Machine and Random Forest classifier. The performance of the trained models varied per patient, feature combination and training algorithm, with the highest accuracy reaching 94%.