Feature Selection Algorithms Based On Grey Wolf Optimization

With the increase of data scale,the difficulty of data mining also increases,and the performance of machine learning in processing high-dimensional datasets also decreases.As one of the important methods of data mining,feature selection can select feature subsets with higher information content and ensure the classification accuracy.At the same time,redundant information can be filtered,data dimensions can be reduced,and performance of subsequent algorithms can be enhanced.In order to study feature selection effectively,this paper will use the meta-heuristic algorithm--Grey wolf optimization algorithm to solve feature selection.The grey wolf optimization algorithm which designed according to the hierarchy and activity behavior of the grey wolf population in the biological world.It can balance the exploration and development process,and show good convergence ability.However,the grey wolf optimization algorithm still has some defects in the process of searching range and optimization when it is used to solve the feature selection.Firstly,for the problem of that grey wolf algorithm is prone to fall into local optimum in the later stage of search,the tabu grey wolf optimization(TGWO)is proposed for this situation in this paper.This algorithm is designed to record multiple historical searches by using tabu table and without changing the update equation of grey wolf population iteration.In the subsequent iteration process,the algorithm can effectively avoid the invalid iteration of grey wolf population,and find out the feature subsets quickly.Secondly,in order to enhance the progeny searching ability of the population,this paper proposes the grey wolf chemical reaction optimization algorithm(CRGWO).Wolf populations in the pursuit of the process of moving prey,we designed five different chemical reactions strategy for wolves,such as the grey wolf redox reaction 1,the grey wolf decomposition reaction,the grey wolf redox reaction 2,the grey wolf wolves replacement reaction and the grey wolf synthesis reaction.This algorithm can update population offspring;generate new gray wolves which have better performance.This algorithm can effectively improve the quality of the next generation of grey wolf population and the convergence rate.To sum up,the two feature selection algorithms are processed by the same transformation function to search two-dimensional features space.The evaluation index includes the fitness,the accuracy and the number of feature subsets,which can give a comprehensive evaluation to the algorithm.The experimental results are compared with meta-heuristic algorithms such as particle swarm optimization algorithm,flower pollination algorithm,whale optimization algorithm,bat algorithm,original grey wolf algorithm and the latest two-stage mutated grey wolf algorithm.Many experiments show that the proposed algorithm has good interpretability and strong selection performance,which is of great significance for solving the problem of feature selection.