Enhanced Epileptic Seizure Detection using Imbalanced Classification

Abstract

Epilepsy is the second most persistent neurological condition, endangering the lives of patients. Though there have been many advancements in neurological imaging approaches, the Electroencephalogram (EEG) still remains to be the most effective tool for testing and diagnosing epileptic patients. The visual analytics of EEG signals is a very prolonged process and always open to the subjective judgment of the physicians. The main goal of our study is to build an automatic classifier that can analyze and detect epilepsy from EEG recordings obtained from epileptic and healthy patients, thus helping the neurosurgeons to diagnose epilepsy in a better way. Synthetic minority oversampling technique (SMOTE) has been used for balancing the EEG dataset and the Principal component analysis (PCA) technique is applied further, for reducing the EEG signal dimensionality. For data classification, seven machine learning classifiers have been used and after comparing the results the authors conclude that Artificial Neural Network (ANN), outperforms the other classifiers by providing an accuracy of 97.82%.