sEMG Signal Analysis using Hand Crafted Features for Detection and Classification of GTC Seizures

Abstract

Seizures with a greater possibility of sudden unexpected death in epilepsy (SUDEP) deteriorates the efficiency of life. Significantly, it is of prime importance to lessen all the factors that are involved in worsening seizures conditions. In this way, life expectancy can be enhanced if the epileptic seizures are timely recognized and detected with the help of efficient applications and devices. Surface electromyography (sEMG) being a non-invasive procedure effectively analyzes the most frequent type of seizures known as generalized tonic-clonic seizures (GTCs). This study proposes a framework that can differentiate between normal (nonGTCs) and seizures activity accompanying its distribution. The signature hand-crafted features from time domain and frequency domain are observed for class recognition of GTC seizures. In the next step, we have classified binary class, abnormal class and multiclass GTC seizures with the assistance of multiple classifiers. Finally, the SVM classifier achieves the accuracy of 97.3% for binary class classification and 92.6% for multiclass classification in GTC recognition. Furthermore, upon feature selection of upmost features based on feature rankings, the SVM classifier with LASSO feature selector attains 98.1% of accuracy for binary class. Also, KNN classifier with Relief-F feature selector gives 95.24% of accuracy for multiclass classification. The acquired results have been refined in contrast with the state-of-art techniques present in literature. This knowledge would help the researchers to evolve adaptable and efficient applications.