Epilepsy activity detection based on optimized one-class classifiers

Epilepsy represents a significant problem which reflects the existence of abnormal and hyper-synchronous discharges in large ensembles of neurons in brain structures. Despite, the epilepsy have been widely studied, its detection in incipient states is still in development. In order to solve this problem using EEG signals, a rigorous classification process have to be made. One-class classifiers are employed due their high performance under unbalanced classes and the lack of available target data from the biosignals but there are several aspects to consider like kernel parameter and the rejection rate parameter related with computational cost and performarce-stability respectively. In this paper it is proposed a methodology to improve the performance of a classification system using a optimized one-class classifer by means authomatic tuning algorithms. The Support vector data descriptor and mixture of Gaussians are used, and their performance and stability are compared, in order to determine the best one-class classifier. To increase the performance, stability and convergency time of the classifiers, the free parameters are optimized by particle swarm optimization(PSO). Using this approach, the sensitivity and specificity have been improve over the 95%. The methodology is tested with a database that correspond to 29 patients with medically intractable focal epilepsies. They were recorded by the Department of Epileptology of the University of Bonn, by means of intracranially implanted electrodes. It provides a new approach in epilepsy detection using EEG signals.