Research On The Application Of Rough Sets And Intelligent Optimization Methods In Feature Selection

With the advent of the information age,the data of all walks of life is increasing explosively,at the same time,the dimension of the data is also climbing.The emergence of large amounts of high-dimensional data has brought serious challenges to the current machine learning technology.As an effective method for processing high-dimensional data,feature selection can not only reduce the computational cost of learning algorithm,but also improve the performance of machine learning model.Feature selection is essentially a combinatorial optimization problem,so using various intelligent optimization methods to solve the feature selection problem has become a research hotspot.In recent years,as two popular swarm intelligent optimization algorithms,artificial bee colony algorithm and Harris hawks optimization have gained a lot of attention.They have been widely used in feature selection because of their advantages such as few control parameters,simple calculation and easy implementation.However,the existing artificial bee colony algorithm and Harris hawks optimization still have some problems,such as low search accuracy,easy to fall into local optimal and poor convergence performance.In addition,there are two conflicting objectives in feature selection: the minimum number of features selected and the highest classification accuracy based on the selected features.Therefore,it is necessary to study how to construct an appropriate fitness function to evaluate the feature subset.In order to solve the above problems,this dissertation combines rough set theory with artificial bee colony algorithm and Harris hawks optimization to carry out the research of feature selection.The main research work and achievements of this dissertation are as follows:(1)Granular rough entropyIn view of the shortcomings of existing information entropy models,this dissertation proposes a new model of information entropy——Granular rough entropy.Granular rough entropy can not only describe the incompleteness of knowledge,but also measure the granularity of knowledge effectively.We prove the monotonicity of grain size rough entropy and some other properties.The proposal of grain size rough entropy will provide a new solution for constructing fitness function of artificial bee colony algorithm and harris hawks optimization.(2)Improved artificial bee colony algorithm based on Granular rough entropy and cloud modelAiming at the problems of the traditional artificial bee colony algorithm,such as low search accuracy,poor convergence performance and easy to fall into local optimum,this dissertation proposes an improved artificial bee colony algorithm based on Granular rough entropy and cloud model IABC＿GRECM(Improved artificial bee colony algorithm based on granular rough entropy and cloud model).Compared with the traditional artificial bee colony algorithm,IABC＿GRECM makes the following improvements: Firstly,aiming at the construction of fitness function in artificial bee colony algorithm,a fitness function based on Granular rough entropy was proposed,and the fitness of each solution was calculated by Granular rough entropy Secondly,in order to improve the local search ability of the artificial bee colony algorithm,the cloud model is introduced into the following bee stage,and the one-dimensional normal cloud model is used to adjust the local search scope.Thirdly,an improved roulette based selection strategy for following bees can avoid premature convergence and ensure population diversity.(3)Feature selection algorithm based on improved artificial bee colony algorithmBased on the improved artificial bee colony algorithm IABC＿GRECM proposed in(2),this algorithm is further applied to the feature selection field,and a feature selection algorithm based on the improved artificial bee colony algorithm FS＿IABC(Feature selection algorithm based on improved artificial bee colony algorithm)is proposed.FS＿IABC uses Sigmoid function to encode the location of honey source and map the features to the honey source.In addition,FS＿IABC uses Granular rough entropy to construct fitness function,which can evaluate feature subsets better.To verify the performance of FS＿IABC algorithm,we conducted experiments on multiple UCI data sets.The experimental results show that the feature selection performance of FS＿IABC is better than the existing feature selection methods based on intelligent optimization.(4)An improved Harris Eagle algorithm based on grain-size rough entropy and gravitational searchAiming at the shortcomings of the traditional Harris hawks optimization,an improved Harris hawks optimization based on Granular rough entropy and gravitational search is proposed in this dissertation.Compared with the traditional Harris hawks optimization,IHHO＿GREGS(Improved harris hawks optimization based on granular rough entropy and gravitational search)makes the following improvements: Firstly,in order to solve the problem of low optimization accuracy of traditional algorithms,Granular rough entropy was used to construct the fitness function,and the inertia mass of Harris hawk was calculated by Granular rough entropy.Secondly,the acceleration calculation method in gravity search algorithm and adaptive step size strategy are used to update the position of Harris Hawk,which can improve the global exploration ability and convergence performance of Harris Hawk.Thirdly,in order to ensure the diversity of the population and avoid the algorithm falling into the local optimal,the adaptive weight method was introduced into the local mining stage of Harris hawk.(5)Feature selection algorithm based on improved Harris hawk algorithmBased on the improved Harris hawks algorithm IHHO＿GREGS proposed in(4),this algorithm is further applied to the feature selection field,and a feature selection algorithm based on the improved Harris hawks algorithm FS＿IHHO(Feature selection algorithm based on improved harris hawks optimization)is proposed.FS＿IHHO uses Sigmoid function to encode the individual state of Harris Eagle,and uses Granular rough entropy to evaluate the feature subsets.To verify the performance of FS＿IHHO algorithm,we conducted experiments on multiple UCI data sets.The experimental results show that the feature selection performance of FS＿IHHO is better than the traditional Harris hawks algorithm and some other representative feature selection methods.