Seizure Prediction in Epileptic Patients Using EEG and Anomaly Detection

Abstract

Epilepsy is a neurological disorder in which abnormal brain activity occurs, causing seizures. Seizures may involve convulsions and loss of consciousness and can harm patients and the people around them. Many patients are drug-resistant, and medication does not improve their situation. Predicting the onset of epileptic seizures may improve their quality of life. For this purpose, many studies have utilized EEG signal, which reflects the brain's electrical activity. This paper contains a new seizure prediction method based on Anomaly Detection and with the help of One-Class SVM. The average sensitivity and False Alarms Rate were 94% and 0.89 per hour. The advantage of this method over other methods, such as classification approaches, is that the network needs much less data for training, as only 8 hours of training data have been used for each patient. Low computational complexity and ease of use make it suitable for real-time prediction compared to other studies.