Comparative Analysis of Feature Selection in Epilepsy Seizure Recognition Using Cuckoo, Gravitational Search and Bat Algorithm

Abstract

While applying classification techniques on high dimensional data, we may not get satisfactory results due to a large number of attributes requiring lots of computational resources, and containing features which aren’t at all related to the response variable, hence making it difficult to be used for practical purposes. To deal with this problem swarm intelligence algorithms are used for selecting the best features for classification. In this paper, we have compared 3 swarm optimization algorithms for their feature selection abilities for recognition of epileptic seizures, which has high dimensional data with 179 features. The base classifiers used were Decision Tree, Random Forest and KNN whose performance was compared with and without feature selection. In our experimentation, we found that for each metric, a different combination of algorithms yielded the best results, for example, random forest with bat algorithm yielded 96.78% accuracy, 96.89% recall value which are the best results among other combinations.