Epileptic Seizure Detection based on Statistical and Wavelet Features and Siamese Network

Epilepsy can be defined, according to the World Health Organization, as recurrent seizures related to physical reactions caused by a sudden discharge of electricity to some human brain cells. Electroencephalogram (EEG) signals play a vital role in the diagnosis of this disease. The recording of EEG signals recorded by mobile recording devices produces very long information that the detection of the epileptic area requires a long time for the expert to analyze all the information. Traditional methods of analysis are tedious, which is why in recent years there have been so many automated systems for diagnosing epilepsy. In this article, a new approach to the diagnosis of epilepsy is presented. First, the preprocessing process is applied to the EEG signals and the signal is decomposed into ten sub-signals using an experimental wavelet transform. Then, the best features are selected using the proposed method of analysis of variance. Then, using the Siamese network to decrease the dimensions of the feature vector in improving the performance of seizure detection. Finally, the support vector machine (SVM) algorithm uses these features to classify convulsive and non-convulsive EEG signals. The simulation results show that the proposed method of the paper using the EEG signal dataset of the University of Bonn has resulted in 99.30 accuracy and this method can effectively help physicians in diagnosing epilepsy, thus reducing their workload.