Research On Adaptive Feature Selection And Parameter Optimization Algorithm For Random Forest

The random forest algorithm(Random Forest)is a popular method of data mining in the field of machine learning.It combines the theory of decision tree algorithm with bootstrap resampling method,and sets up multiple single tree classifiers.The final result is classified and predicted by voting strategy.Random forest algorithm has many advantages,such as easy to understand and adjust parameters,strong noise immunity,and most importantly,in practical applications,it has very high classification performance and is not easy to be overfitted.The properties of good random performance and the lack of sample background knowledge have made it widely used in many fields.For this reason,many researchers have conducted extensive research and improvement on random forests.This paper analyzes relevant research at home and abroad and finds that random forest algorithm's feature selection is random,which leads to the uncertainty of decision tree training accuracy.At the same time,random forest parameters also have limitations that are difficult to select.In view of the above issues,this article has done a series of exploration and research work on the selection of features and parameters of random forests from different perspectives.This paper first describes the review of random forest algorithm,and analyzes the stochastic mechanism,performance index and problems of random forest algorithm.Then,according to the randomness of random forest feature selection,an adaptive feature selection classification algorithm named SARFFS is proposed.The algorithm first uses Chi-square to test the degree of association between samples and then self-sampling,and designs a feature to calculate the strength of class representation.Method;Then the adaptive sparse constraint mechanism Group LASSO optimization feature selection is introduced to solve the limitations of the random forest selection feature.Finally,an improved particle swarm optimization algorithm is proposed to optimize the random forest parameters for the subjective disturbance of random forest parameter selection.Selecting,this method first improves the particle swarm from the learning factor of the particle swarm and the position of the particle.For different requirements of different phases of particle motion trajectory,a strategy based on the inverse sine adjustment factor is proposed based on the learning factor.When the traditional PSO algorithm updates the position of particles,it does not consider theinfluence of context on the particles during each iteration.This paper updates the recursive equations of the PSO algorithm iteratively and proposes a new recursive formula.Algorithm optimization capabilities.