A Survey on Application of Metaheuristics Techniques for Ensemble Feature Selection (EFS)

Abstract

Unprecedented increase in the quantum of data generated through online social, financial, business and learning transaction have necessitated the incorporation of indigenous techniques to intelligently preprocess/trim the data to a level that suits the learning algorithm in all dimensions with improved classification accuracy and performance. Intelligent Feature Preprocessing techniques are deployed to address the issue of "curse of dimensionality" that is persistent in the current voluminous databases. Individual feature selection techniques when applied to the dataset may face several deficiencies viz. may get stagnated in the local optima of the complex feature search space, tends to yield unstable feature subsets missing the feature interaction etc. These challenges have encouraged the implementation of Ensemble Feature Selection (EFS) techniques where multiple feature selection models (Base Feature Selectors) are integrated to form a composite model (stronger version) to yield an optimal and robust feature subset. Eventual application of these features to the learning/classification algorithm produces better results with high accuracy and efficiency. High quantum of feature subsets generated through EFS necessitates the replacement of conventional feature searching/ranking/combining strategies with stochastic metaheuristic techniques exclusively designed for tackling high dimensional data issues to generate a global optimal solution. The voluminous intermediate feature subsets from the EFS mandates the deployment of Population based multi-objective Metaheuristics algorithm viz. Minimizing the number of features, classification error, minimizing the total misclassification cost and maximizing the classification accuracy. Population based Metaheuristics techniques serves as correct fit for seamlessly finding optimal feature subsets that sieves through multiple feature subsets generated through EFS. This paper aims to present a survey encompassing the adoption of metaheuristics techniques for efficient feature selection with an aim to improve the detection rate of various attack classification accuracy and efficiency.