Feature Evaluation for Seizure Prediction Using EEG

Epilepsy is a brain disorder with which more than 65 million people suffer worldwide. It is still a willful challenge without a definitive solution, regardless of extensive research. With the descent of these seizures, a spark developed in researchers to focus on predicting epileptic seizures via electroencephalogram (EEG). The advancement in machine learning in the field of medicine allows the researchers to bring off the seizure prediction algorithm accurately. In this work, the CHBMIT database for predicting seizures is used. The data set of five patients has been taken with seizure and seizure-free recordings. The raw data is pre-processed, and several features have been extracted. After analyzing the discriminatory nature of different features, these features were fed to various Machine Learning Classifiers to classify pre-seizure from normal states and Regression trees classifier outperformed all the other classifiers achieving accuracy of 93.3% with 87% true positive rate (TPR).