改进的K-means融合微粒群优化的基因选择方法

Gene Selection Method Based on the Fusion of Improved K-means Algorithm and Particle Swarm Optimization

为了解决基因选择困难问题,提出一种基于改进的K-means算法融合微粒群优化(IKPSO)的基因选择方法。该方法首先运用过滤法(Relief)对基因进行筛选,选择出对分类贡献大的基因构成备选基因子集;然后,利用改进的K-means算法将备选基因子集划分为一定数目的簇,并运用微粒群(PSO)对每一类簇进行搜索选择出相应类簇中的最优和次优基因构成最优特征基因子集;最后,训练支持向量机(SVM),并利用其分类的性能来评价获得的最优特征基因子集的质量。在两个典型的、公开的小样本的高维微阵列数据集上进行的实验,结果表明该IKPSO算法总体分类性能相对较好,并且与传统方法相比,IK-PSO分类性能得到显著的提高,证明了IK-PSO的可行性以及有效性。

In order to solve the difficult problem of gene selection,this paper proposes a gene selection method based on the fusion of improved K-means algorithm and particles warm optimization（IK-PSO）.Firstly,this method uses filtration（Relief） to screen genes and selects genes that contribute greatly to classification to form subsets of candidate genes.Then the subsets of candidate genes is divided into certain number of clusters by the improved K-means algorithm,and using the particles warm（PSO） to search every cluster to select the optimal and suboptimal genes in the corresponding cluster to form optimal subset of feature genes.Finally,the support vector machine（SVM） is trained and the quality of the obtained optimal feature subsets is evaluated by its classification performance.The experiment results on two open,typical small sample of high-dimensional microarray data sets showthat the overall classification performance of the IK-PSO algorithm is relatively good,and that the IK-PSO classification performance is significantly improved compared with the traditional methods,and the feasibility and availability of the IK-PSO are also proved.