Machine Learning Framework for Identification of Abnormal EEG Signal

Abstract

Epilepsy is a chronic, noncommunicable disease (NCD) causing disorder in the brain’s neurological activity. It may be due to genetic disorder or brain injuries caused by some accidents. This may cause seizures, loss of awareness, unusual sensations and behavior. Globally 50 million people are suffering from epilepsy, so it is one of the most prominent neurological diseases globally according to World Health Organization (WHO) statistics in 2021. It is estimated that up to 70% of people are suffering with epilepsy and they can be saved by timely diagnosis and proper treatment of epilepsy. Electroencephalograms (EEGs) are universally used to detect this chronic non-communicable disease. Furthermore, assessing a specific type of abnormality by visual examination of an EEG signal is an intuitive process that can vary from radiologist to radiologist. It is a challenging task for the radiologists to visually examine the EEG signal by looking for a shift in frequency or amplitude in long-duration signals. It may give rise to inaccurate categorization. Identification of epileptic seizure from the recorded EEG signal is a primary task in the treatment of epilepsy. In this work, wavelets were used to obtain the appropriate features from EEG signals. These features were fed to different classifiers. This work proposes a machine learning (ML) framework to detect the abnormality in the EEG signal automatically to assist the radiologists in their diagnosis. The ML framework uses 7 classifiers (KNN, SVM, Random Forest, Logistic Regression, Decision Tree, AdaBoost, and Bagging). Among these classifiers, Bagging Classifier was shown better performance in terms of accuracy and ROC.