基于粒子群优化算法的最大相关最小冗余混合式特征选择方法  

A maximum relevance minimum redundancy hybrid feature selection algorithm based on particle swarm optimization

作　　者：姚旭[1] 王晓丹[1] 张玉玺[1] 权文[1] 

机构地区：[1]空军工程大学防空反导学院,西安710051

出　　处：《控制与决策》2013年第3期413-417,423,共6页Control and Decision

基　　金：国家自然科学基金项目(60975026;61273275)

摘　　要：在分析粒子群优化(PSO)算法和简化PSO算法的基础上,提出一种基于PSO的最大相关最小冗余的Filter-Wrapper混合式特征选择方法.Filter模型是基于互信息和特征的相关冗余综合测度,Wrapper模型是基于改进的简化粒子群算法.在PSO搜索过程中,引入相关冗余度量标准来选择特征子集,将Filter融合在Wrapper中,利用Filter的高效率和Wrapper的高精度提高搜索的速度和性能.最后以支持向量机(SVM)为分类器,在公共数据集UCI上进行实验,实验结果表明了所提出算法的可行性和有效性.A Filter-Wrapper hybrid feature selection approach with maximum relevance and minimum redundancy based on particle swarm optimization（PSO） algorithm is proposed on the analysis of PSO algorithm and simplified PSO algorithm. The Filter is based on mutual information and the composite measure of feature relevance and redundancy, while the Wrapper is based on a simply modified PSO algorithm. The relevance and redundancy criterion is introduced to select features in the PSO＇s searching procedure. Meantime, the Filter is fused into the Wrapper. The speed and performance of the search are improved with the higher efficiency of the Filter and the greater accuracy of the Wrapper. The experiment results based on UCI data sets with support vector machine（SVM） as the classifier show the effectiveness and feasibility of the algorithm.

关 键 词：特征选择 粒子群优化 FILTER WRAPPER 互信息 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]