Epileptic Seizure Detection Using Machine Learning Techniques

Abstract

Epilepsy is a neurological disorder, which causes seizures. Detection of epilepsy is carried out by analyzing EEG signals. Detecting epileptic seizures from long-term EEG data is a time consuming and tedious task, which requires vast clinical expertise. Most of the epilepsy detection algorithms available today are highly patient dependent. In this paper, an efficient and patient independent epileptic seizure detection algorithm based on machine learning is proposed. We have developed a method to classify seizure and non-seizure data using different machine learning algorithms. Time-domain, frequency-domain, and wavelet-domain features are used in this work. Feed-forward neural network, anomaly detection using multivariate Gaussian distribution and long short-term memory network are employed to classify seizure and non-seizure data. CHB-MIT database is used in this study. Long short-term memory network has given the highest seizure detection accuracy (97.4%) and the lowest false positive rate (7.88%).