A Novel Differential Selection Method Based on Singular Value Decomposition Entropy for Solving Real-World Problems

Abstract

An evolutionary singular value decomposition (SVD) entropy based feature selection approach is proposed for finding optimal features among large data sets. Since the data typically consists of a large number of features, all of them are not optimal. In this paper, an optimal feature selection approach based on differential evolution (DE) and SVD entropy is proposed. The functioning of the proposed approach is examined on available UCI data sets. This approach provides ranked features by optimizing SVD entropy using the DE. An SVD entropy based fitness function is employed as the criterion to measure the optimal features and this makes the new approach easier to implement. DE results in a faster and accurate convergence towards global optima. The proposed approach shows its effectiveness on binary data sets with a number of features ranging between 9 and 60. The result explains that the proposed approach can converge quickly and rank the features. The experimental section demonstrates the results in terms of classification accuracy by Support Vector Machine (SVM) and Naive Bayes (NB) classifiers. The explored results are favorable and strengthen the contribution of the proposed approach.