Preictal phase detection on EEG signals using hybridized machine learning classifiers with a novel feature selection procedure based GAs and ICOMP

Abstract

Epilepsy is the fourth most common neurological disorder, which affects the brain and brings out frequent seizures. They are bursts of electrical discharge that can cause a wide range of symptoms such as distraction or involuntary spasms involving the whole body. Preictal phase carries some important features related to seizures which can be found before seizure onset. Hence, this study introduces an efficient hybrid training procedure for machine learning (ML) classifiers that are able to classify Electroencephalogram (EEG) signals for the accurate detection of preictal phase. Essentially, the proposed approach consists of two stages: feature extraction and model estimation with feature selection. In this approach, while the feature extraction is executed by using wavelet transform, the model estimation is performed by hybrid ML classifiers. Essentially, this approach integrates the training mechanism with a novel feature subset and model selection procedure based on the Information Complexity Criteria (ICOMP) and Genetic Algorithms. For preictal phase detection application, the CHB-MIT Scalp EEG dataset was analyzed by both the proposed and traditional approaches. From the analysis results, it can be concluded that the hybrid ML classifiers not only produce robust models in the context of model information complexity, but also provide superior performance outputs than the classical approaches with respect to validity and reliability, over test datasets.