Research On Feature Selection Based On Improved Forest Optimization Algorithm

With the continuous development of data mining and machine learning in the field of computer science,facing the growing scale of data sets,how to maintain excellent algorithm performance on the basis of reducing computation cost has become a crucial issue.Feature selection which goal is to make selected subset got as much as or even better than the performance of the original data set in the algorithm,meanwhile,to reduce the dimension of data reasonably.In the classification algorithm,in order to get better classification performance,the unrelated and redundant features will be removed from the original feature set.Feature selection method selects the most representative feature from the large initial feature set,and adds it to the feature subset.There are numerous ways for feature selection.In recent years,the feature selection by evolutionary computation has gained wide attention in academic circles,and has achieved remarkable results.Forest Optimization Algorithm is an evolutionary computation method that simulates the method of planting trees in the natural world.Initially,it is used to solve the continuous optimal problem.After optimization by scholars in machine learning,the FOA evolved into the FSFOA(Feature Select using Forest Optimization Algorithm)to solve the feature selection problem.In this paper,we propose SFSFOA—Strengthen Feature Selection using Forest Optimization Algorithm based on the FSFOA algorithm,aiming at the shortcomings of FSFOA algorithm.SFSFOA proposes three optimization strategies,namely,Strengthen Seeding,Hasten Grow and Dominance Tree Hybridization.SFSFOA algorithm is experimentation on the data sets of three dimensions of low and medium high.It is proved that the SFSFOAalgorithm not only has a further improvement in accuracy,but also has considerable improvement in dimension reduction.The classification method of feature selection can be divided into Filter and Wrapper according to the evaluation criteria.The Filter method compared to the Wrapper method has the advantage of small computational complexity,low computational cost,high efficiency in high-dimensional data;and the Wrapper method to study the results oriented,higher classification performance.SFSFOA algorithm is the same as FSFOA,which belongs to Wrapper method.And it also has the disadvantage of high cost of Wrapper method.Therefore,a WFFSFOA(Wrapper and Filter based on Feature Select using Forest Optimization Algorithm)algorithm is proposed on the basis of SFSFOA.In WFFSFOA,Filter and Wrapper — two types of feature selection methods are integrated.Compared with the FSFOA algorithm,WFFSFOA can not only guarantee the accuracy of the algorithm,but also reduce the calculation cost significantly,and improve the convergence performance of the algorithm.