Research On Epileptic EEG Signal Classification Based On Particle Swarm Optimization And RBF Neural Network

Epilepsy is one of the common diseases and frequently-occurring diseases which endangers the health of human beings and gets more attention in many countries.EEG is an important method used to detect the epileptic.However the EEG of epileptics don't show abnormal all the time,it still exists problem to recognize the epileptic depending on the EEG only.Moreover,researchers have found that the randomness,instability and nonlinearity of epileptic cause much trouble to the further study.Consequently,the effective way to extract the EEG feature for presenting the information of epileptic EEG is the priority of epilepsy diagnosis.To deal with the randomness,instability and nonlinearity of epileptic,this dissertation proposes a method to classify epileptic EEG.Firstly,we extract the feature via time domain and nonlinear analysis approach and then optimize the feature selection by Particle Swarm Optimization algorithms(PSO).Lastly we train RBF neural network to build the epileptic classifier using the extracted features after optimization.In the experiment,we analyze the extracted features on the categories of epileptic.And then we apply PSO to optimize the selected features when doing the binary-classification and three-element classification.It shows that the optimized result matches the features analysis by boxplot,especially the precision of binary-classification and three-element classification can reach 99%and 98.1%.The results of various cross validation experiments demonstrate that our method can be very effective in extracting the feature in epileptic EEG classification.This paper also proposes an effective method in EEG framing,which employs the features generated by features selection optimization model.This method can detect the epileptic EEG,and improve the accuracy of the classification.