Research On Feature Selection Methods Based On Swarm Intelligent Algorithm

In today's era of big data and intelligent development,feature selection methods have always been a key link in data mining,machine learning and other fields.Usually a model building and data analysis process spend most of the time on data preprocessing,which hinders the research process to a certain extent.Therefore,intelligent feature selection simplifies data preprocessing.Cheng is a research direction.However,a single intelligent algorithm may be limited by inadequate accuracy,slow convergence speed and local optimal solution.Therefore,it is of great significance to consider the fusion of multiple intelligent algorithms and apply them to the research of feature selection methods so as to achieve the effect of complementing each other's strengths and improving the performance of the algorithm.This paper mainly includes the establishment of one-dimensional classifier based on random forest,the research of feature selection based on genetic algorithm and the research of weight optimization based on particle swarm optimization.With the help of the idea of Stochastic Forest democratic consultation,each sample is voted to determine the classification method of the corresponding features,and the optimal threshold is determined by traversal method for binarization,so that each feature is trained as the optimal one-dimensional classifier.With the idea of evolutionary strategy and thescreening principle of "natural selection of things,survival of the fittest",the combination mode of features in each evolutionary process is adjusted through selection,crossover and mutation mechanism,and the fitness function is taken as the goal,and the classification features are finally screened out in the process of continuous reproduction.Since fitness is the only evolutionary goal of this paper,the features selected by this method are easy to handle and objective.In the process of predation,birds have the mechanism of memory and sharing information to optimize the weight.Each group of weights is regarded as a particle,and multiple groups of different weights constitute the particle swarm.Based on the iteration of their local optimal solution and the global optimal solution of the whole group,the optimal fitness value is approached continuously.Finally,this paper finds that swarm intelligence optimization algorithm simulates the natural biological interaction,system action and continuous evolution.Compared with traditional optimization algorithm,it has the advantages of simple form,no complicated mathematical derivation and easy implementation.Combining genetic algorithm and particle swarm optimization algorithm,it has better effect on solving the system problems of feature selection and weight optimization,and greatly reduces the cost.The process of feature selection is cumbersome.