An Information Entropy Embedding Feature Selection Based on Genetic Algorithm

Abstract

Feature selection is of vital importance to reduce information redundancy and deal with the invalidation of basic classification approaches for massive dataset and too many features. In order to improve the classification accuracy and decrease time complexity, an algorithm with intelligent optimization genetic algorithm and weight distribution based on information entropy is proposed, called EEGA. Information entropy of features is defined as the population labels in GA rather than the direct iteration of individual fitness. Experiments have been performed by using several standard databases with four fitness algorithms. Experimental results have proved that EEGA performs better based on the measure of accuracy. Furthermore, it can significantly reduce the required time when figuring out the better results.