With the popularization to computer and Internet,the rapid development of database technology and the related data acquisition and data storage,data mining playing an more and more important role in data analysis in every industry.With the further development of Computer Science,the Age of Big Data is coming.Meanwhile,large high-dimensional data are emerging,and these high dimensional data exists in a lot of redundant,irrelevant characteristics.It has brought great challenge to data mining technology.At this point,dimension reduction is very necessary.Feature selection,as a dimension reduction methods,has also become a research hotspot in data mining and develop quickly.This paper introduces the general process,search strategies and optimization methods of feature selection.And it has analyzed of research status and research direction of feature selection.According to advantages of the support vector machine(SVM)classifier,the SVM has been used for feature selection.And an improved Gravitational Search Algorithm(GBGSA)was proposed for feature subset search and SVM parameters optimization.Then constructing a system that based on GBGSA and SVM.Firstly,this paper introduces the theory of SVM and several swarm intelligent algorithms.And it introduces how the swarm intelligence algorithms been used for SVM parameters optimization and feature selection.Secondly,this paper describes Gravitational Search Algorithm(GSA)and binary Gravitational Search Algorithm(BGSA)in detail.,To overcome the lack of local optimization capacity and the problem of premature search convergence of Gravity Search Algorithm,an improved Gravitational Search Algorithm(GBGSA)was proposed.Crossover and mutation operations were introduced from Genetic Algorithm.Using tournament selection methods of the particle swarm,proposed a method of Adaptive mutation to mutation particle transfer speed.Then based on a benchmark function test,the effectiveness of the algorithm is verified.Finally,based on improved Gravitational Search Algorithm,simultaneous optimizing SVM parameters and feature selection,build the GBGSA-SVM system.According to experiments on UCI data sets,compare to genetic algorithm and search algorithm of gravity results,the optimality of the improved algorithm is verified.And GBGSA-SVM is applied to cataract risk factors analysis,and providing some reference value for pathological study of cataract and prevention. |