Application Of Improved Particle Swarm Optimization Algorithm In Feature Selection

With the rapid development of various industries,the explosive growth of massive data has brought great pressure to data processing.This also puts forward stricter requirements for data processing methods,which requires more efficient reduction of data dimension and pressure in the process of data use.In recent years,the problem of feature selection has attracted more and more attention.Because of the increase of data dimension and data quantity,the traditional exhaustive method has been unable to properly solve the problem of feature selection(FS).Therefore,many scholars have made continuous improvements and improvements in feature selection methods,and achieved considerable results.Swarm intelligence optimization algorithm(SIOA),as a heuristic method,has the advantages of intelligent selection and random search,and can search ideal solution set under the premise of limited computing resources and time cost.Particle Swarm Optimization(PSO),as a typical representative of SIOA,is also very suitable for solving FS problem under the background of current data explosion because of its fast convergence speed and small parameter impact.Therefore,it is of great practical significance to analyze the principle of PSO and put forward further improvement,so that it can be applied to solve various practical problems including FS.The performance of PSO algorithm is analyzed.Firstly,PSO algorithm has the advantages of fast convergence and small parameter dependence.Therefore,it can quickly converge to the optimal solution region at a lower cost.On the other hand,PSO algorithm still has obvious defects.Therefore,based on the density of the particles in the decision space and the compactness in the solution space,this paper improves and improves the diversity of the population respectively.At the same time,considering the specificity of FS.several improved strategies are put forward to apply in the field of FS problem.Finally,the performance of the improved algorithm is verified by the classification performance test on the actual data set.The improvement of this paper is mainly divided into the following parts:(1)Based on the Binary Particle Swarm Optimization(BPSO)algorithm,a strategy of population initialization based on feature clustering information is proposed.Secondly,an adaptive local search strategy based on decision space similarity is proposed,in which the similarity index of particles is determined by the similarity of particles in decision space.In the process of population evolution.according to the similarity of particles in decision space,the local search force is adaptively adjusted,so that the particles in the population can be adaptively disturbed to avoid premature BPSO algorithm.Finally,three representative optimization algorithms are selected to carry out comparative experiments on 11 UCI datasets.The experimental results show that the improved BPSO algorithm has a better performance in reducing the number of features than other comparison algorithms,and the classification accuracy is also significantly improved.(2)By analyzing the framework of BPSO algorithm,its single learning model easily leads to the limitation of particle learning.Then a hierarchical learning model based on the density of particles in solution space is proposed.During the iteration process,the particles are stratified by the compactness of fitness values,and then the advantages of each layer of particles in solution space are brought into full play,and a more reasonable learning model is selected.Considering the stability,search performance and convergence performance of the algorithm,the limitation of the algorithm itself is avoided.Finally,through the comparative experiments on 10 datasets,it is verified that the diversity of learning models can improve the performance of the algorithm and get a better feature subset.