A Quasi-Oppositional-Chaotic Atom Search Optimization Algorithm to Detect Epileptic Seizure from EEG Signal Using WPT and ELANFIS Classifier

Epilepsy is a complex neurological disease that causes seizures and happens due to the sudden flow of electrical signals from the affected brain part of a healthy person. Although it is not that harmful, in some cases, it seizes a person's everyday activities in such a way that the person may lose consciousness, feel tired, and nervous, and, in most cases, face accidents due to memory failure. To keep up with all these dangerous incidents, attempts have been made in neurological science to develop methods to detect the symptoms of seizures. Because epilepsy is not curable, so if detection can be accurate, then preventive steps can be taken it escape from the seizure. Machine learning in the search process integrated with metaheuristic optimization methods is the most efficient approach developed by researchers to classify seizure symptoms. In this study, a new patient-specific real-time machine learning model for automatic epileptic seizure detection in which wavelet packet transform (WPT) is implemented for feature extraction and Extreme Learning Artificial Neuro-Fuzzy Inference System (ELANFIS) is used for classification. Quasi-oppositional based chaotic learning is integrated with atom search optimization algorithm to improve the search exploration and exploitation. The whole experimental analysis is employed using the Bonn University EEG dataset. Measurement parameters obtained from proposed algorithms as average accuracy, sensitivity, specificity, Positive predictive value, Mathews correlation coefficient, and Area under curve (AUC) are 99.8%, 99.35%, 99.82%, 96.85%, 98%, and 1. The proposed algorithm is outperformed by the conventional ASO algorithm. In the future, other optimization algorithms and IoT will be implemented for advanced clinical applications.