Energy-aware cross-layer optimization for EEG-based wireless monitoring applications
Epileptic detection techniques rely heavily on the Electroencephalography (EEG) as representative signal carrying valuable information pertaining to the current brain state. For these techniques to be efficient and reliable, a set of discriminant, epileptic-related features has first to be obtained. Furthermore, depending on the classifier model used, a subset of these features is identified and selected for the classifier to yield an optimum performance. Many feature extraction and selection techniques have been reported in the literature, utilizing different strategies. The aim of this work is to review the most widely used ones and to evaluate their performance in terms of their overall complexity and classification accuracy. For this purpose, the support vector machine (SVM) is chosen as a classifier model to study the performance of the obtained features. Extensive experimental work has been carried out and the comparative results and trade-offs are reported.