Towards a better multivariate time-series detection of epileptic seizures in electroencephalogram (EEG) using Machine Learning algorithms

Abstract

This study aims to improve the automatic detection of epileptic seizures (ESs) using machine learning (ML) algorithms applied to electroencephalography (EEG) brainwave data. Previous studies based on a database published online showed high seizure detection accuracies, but they contrasted seizure activity to all kinds of non-seizure EEG activity, recorded from different populations (healthy and patient) and different types of electrodes (surface and intracranial). Here we decided to focus on detecting seizures from non-seizure activity recorded from the same type of electrodes in the same group of patients. We applied different ML classifiers such as Extreme Gradient Boosting (XgBOOST), Naive Bayes (NB), k-Nearest Neighbor (k-NN), Random Forest (RF), Support Vector Machine (SVM), Linear Regression (LR), and Decision Tree (DT). The best Area Under Curve (AUC) value is given by XgBOOST with 95.84%. This research helps to improve the detection of human ES in EEG signal recordings.