EEG Classification for Seizure Prediction Using SVM vs Deep ANN

Epilepsy is a brain disorder in which abnormal brain activity occurs, causing seizures. Epilepsy causes sudden seizures which affects patients medical and social life. Prior investigations for seizure treatment utilized seizure prediction to offer a solution for epileptic seizures. The past two decades witnessed several advances in classification of electroencephalogram (EEG) to assess in seizure detection. Advancements in this field include in-vivo devices that measure brain neural activity, however this damages brain tissues and has an irreversible effect. In this paper, the major problem of seizure classification for drug-resistant epileptic patients using two different algorithms is presented. Two cascaded classification models, SVM and deep ANN, are presented and implemented. The models were created in need of both feature extraction and classification of both pre-ictal and inter-ictal EEG signals. In addition, high precision and accuracy was a main thought in requirement to show that these classifiers are effective compared to other trials in such path. The aim was fairly reached with an accuracy of 97% for the first classification method and 96.7% with cross validation, whereas the second classification model resulted in a accuracy of 92% and 92.3% with cross validation.