A new wrapper feature selection model using Skewed Variable Neighborhood Search with CE-SVM algorithm

Abstract

Feature selection is an important step in many Machine Learning classification problems. It reduces the dimensionality of the feature space by removing noisy, irrelevant and redundant data, such that classification accuracy is enhanced while computational time remains affordable. In this paper, we present a new wrapper feature subset selection model based on Skewed Variable Neighborhood Search (SVNS). In order to determine classification accuracy, we endorse Support Vector Machine (SVM) which is a well tested classification algorithm. The optimal feature subset is investigated using SVNS while SVM hyperparameters are automatically tuned by Cross Entropy (CE) technique which is recognized to be a powerful optimization tool. The performance of proposed model is compared with some existent methods regarding the task of feature selection on 3 well-known UCI datasets. Simulation results show that the suggested system achieves promising classification accuracy using a smaller feature set.