Search

Epilepsy is a complex process and the prediction of seizure is uncertain. Currently, many researchers are exploring different optimization methods to detect the epileptic seizure. Atomic search optimization (ASO) shows a better search capability by using interaction force and the constraint force but it still some deficiency from a local optimum and a low search efficiency. To overcome these demerits, a new enhanced search ability based ASO named as ESAASO is proposed. In this study, inertia weight, levy flight and ranking strategies are integrated into ASO to improve the search performance. The proposed method has been studied on CHB-MIT scalp EEG database. There are 13 number of features are extracted by using TQWT and features are selected using genetic algorithm (GA). The proposed detection method has been computed using twelve well known optimization algorithms with LSSVM classifier. We have obtained average accuracy, sensitivity, selectivity, specificity, average detection rate, G-Mean and area under curve (AUC) values as 98.37%, 91.11%, 91.67%, 91.46%, 91.28%, 91.28% and 0.992 respectively using 10-fold cross-validation method. The proposed algorithm is found outperform as compared to conventional ASO for seizure and non-seizure detection. The proposed method can be experimented using short term EEG datasets in future.