Epileptic Seizure Detection using Empirical Mode Decomposition and Heterogeneous Features with Machine Learning

Abstract

Many people around the world are suffering from epilepsy, a common neurological disorder. Electroencephalograms (EEG) are the primary diagnostic tool, but traditional manual analysis is time-intensive and prone to errors. Various automatic approaches have been proposed for epilepsy detection and monitoring from EEG. However, several challenges, such as non-stationary and nonlinearity of EEG signals, and interand intra-variability, limit the performance of those approaches, especially in real-time applications and e-healthcare. This paper presents an automated approach for epileptic seizure detection using Empirical mode decomposition (EMD), heterogeneous features, and machine learning methods to overcome these challenges. It employs EMD to decompose EEG recordings into a set of IMF (intrinsic mode functions). Thereafter, several heterogeneous features are extracted from different IMFs, representing the EEG patterns. These features are then used to feed machine learning classifiers, using cross-validation scheme. The performance of our framework is evaluated on the publicly available Bonn University EEG dataset. The experimental results reveal that the classification accuracy reaches $\mathbf{9 8. 4 0 \%}$ for ABCD-E.