Research On Attribute Reduction Of Rough Set Based On Social Spider Optimization

With the development of society and the progress of science and technology,new technologies such as data mining and machine learning have been integrated into our daily life,making life more convenient.However,as the amount of data increases,data types become more complex,which limits these technologies.Therefore,how to remove redundant data from the data,ensure data simplification,and provide convenience for subsequent researchers to use these data for analysis is particularly important.Rough Set Theory(RST)is a data mining technology.The most important feature is that it can process inaccurate,incomplete and uncertain data without any prior information.Attribute reduction(AR)is one of the important contents of rough set theory,and it is also one of the data mining preprocessing steps.Its main purpose is to ensure the classification ability and remove the unimportant attributes from the data.Selecting the minimum cardinality from all the reductions is also called minimum attribute reduction(MAR).How to obtain the minimum attribute reduction has been the focus of research.However,solving MAR has been proved to be an NP-hard nonlinear combinatorial optimization problem.In this paper,it is difficult to solve the minimum attribute reduction using traditional attribute reduction algorithms.Intelligent optimization algorithm can make up for the deficiency of traditional attribute reduction algorithm.Therefore,the minimum attribute reduction is solved by combining the social spider optimization algorithm in the intelligent optimization algorithm,but in the actual solution process,the result is often non-minimum attribute reduction and the solution time is long.For this reason,several improvement strategies are proposed,and two algorithms are designed,one is multi-strategy improved social spider attribute reduction algorithm(MISSOAR)with higher solving quality,the other is quick social spider optimization attribute reduction algorithm(QSSOAR)with shorter running time.The research work in this paper is as follows:Solve the minimum attribute reduction using the unique advantages of social spider optimization.First,an opposition-based learning initialization scheme was introduced to improve the performance of individuals in the initial population.Second,the population dynamics were divided into elite population and general population,and an improved female spider location update operator was designed.Then,to avoid the failure of mating and substitution,cross-mutation strategies were introduced to compensate.Finally,a redundancy detection mechanism based on attribute importance was proposed,which can simplify the reduction set by determining the importance of attributes in the optimal results in turn and eliminating the unimportant attributes.The experimental results show that the quality of reduction results is high.Consider that social spider optimization takes a long time to solve the minimum attribute reduction problem.First,the initial population is constrained by a similarity constraint strategy,which requires less modulo-binary addition than the opposite initialization constraints.In the iteration process,an adaptive opposition-based learning was designed to speed up the convergence rate,which is less computational than the traditional opposing learning.In-depth analysis of social spider optimization and other intelligent optimization algorithms in solving the minimum attribute reduction process produces a large amount of computational resources,considering that the maximum computational cost comes from the stage of individual evaluation,a quick extraction strategy was designed.By combining arrays with key-value pairs,the corresponding values can be obtained by simple comparison.The experimental results show that the algorithm has high quality and short running time.