改进的离散PSO和SVM的特征基因选择算法

针对现有的基于粒子群的特征基因选择算法易于陷入局部最优的问题,提出了一种改进的离散粒子群和支持向量机的特征基因选择算法IDPSO-SVM.该算法首先预选一些与分类强相关的基因组成特征基因备选集合,然后基于此集合采用PSO进行寻优搜索,并应用SVM对选出的特征子集的分类能力进行评估,最后得出最优特征子集.该算法加入了一种可以有效克服粒子群在寻优过程中陷入局部最优的机制,因而可以不断探测到新的最优解.该算法在结肠癌与前列腺癌数据集上的分类精度分别达到了96.8%与99.0%,从而证明了其有效性与可行性.

Feature gene selection by combining an improved discrete PSO and SVM

Most existing particle swarm-based feature gene selection approaches have a common problem, easily sinking into a local optimum. To solve this problem, a novel feature gene selection approach combining improved discrete particle swarm optimization and support vector machine (IDPSO-SVM) was proposed. The process began with extraction of genes with closely related classifications. Then particle swarm optimization (PSO) was used to search feature gene subsets in those extracted genes. The support vector machine (SVM) was then used as an evaluator to estimate performance of the selected feature gene subsets. The optimum feature gene subset was eventually acquired after massive iterations. It was noteworthy that the proposed algorithm uses a novel mechanism which may effectively avoid local optimums so that new optimum solutions are constantly found. Classification accuracy of the proposed algorithm on colon tumor dataset and prostate cancer dataset was respectively 96.8% and 99.0%, significantly higher than previous methods. This showed the proposed algorithm is effective and feasible.