Efficient Prediction and Classification of Epileptic Seizures Using EEG Data Based on Univariate Linear Features

Epilepsy is defined as seizures which happen due to disorder in brain functionality. It is certain kind of seizures which effect patient in more than twice in a day when patient loss its senses completely or partially for a short duration. Mostly people affected from epilepsy live in less developed or developing countries. Diagnosis of epilepsy at early stages is quite useful for better treatment of patients. Normal method of diagnosing epilepsy is to admit the patient into hospital and by viewing its EEG recordings. This method is not useful as this include viewing of EEG signals for many hours. In our paper, we propose an algorithm by using which we are able to predict epileptic seizure. There are three states of seizure that include pre-ictal state which is before the start of seizure, another is ictal state during which seizure is happening and there is also a post-ictal state which is after seizure. We have proposed an algorithm by using which we can predict seizure of affected patient i.e; pre-ictal state. We have applied our algorithm on publically available EEG dataset and it has been observed that average pre-ictal time is 34 minutes. It means that we are able to predict epilepsy 34 minutes on average before it actually starts. In this way, there is a sufficient time for medical specialists to start medication in order to avoid seizure. We have also classified the EEG data of patients during ictal state and it has been observed that true positive rate (TPR) is 88.90%.