基于对称不确定性和SVM递归特征消除的信息基因选择方法

发布时间：2018-03-31 22:01

本文选题：基因选择　切入点：对称不确定性　出处：《模式识别与人工智能》2017年05期

【摘要】：基因表达谱中存在大量与肿瘤分类无关的基因,严重降低肿瘤诊断的准确率.基因表达谱还存在高维小样本、噪声大等问题,增加肿瘤诊断的难度.为了获取基因数量较少且分类能力较强的信息基因子集,文中提出基于对称不确定性(SU)和支持向量机递归特征消除(SVM-RFE)的信息基因选择方法.首先利用SU评估基因和类标签之间的相关性,根据SU定义近似马尔科夫毯,快速消除大量无关和冗余基因.然后利用SVM-RFE进一步剔除冗余基因,获取有效的信息基因子集.实验表明,文中方法可以在选取维数较少或相等的信息基因子集情况下获取较高的肿瘤分类性能.
[Abstract]:There are a large number of genes not related to tumor classification in gene expression profiles, which seriously reduce the accuracy of tumor diagnosis. There are many problems in gene expression profiles, such as high dimension and small samples, large noise, etc. In order to obtain a subset of information genes with fewer genes and stronger classification ability, In this paper, an information gene selection method based on symmetric uncertainty (SUU) and support vector machine recursive feature cancellation (SVM-RFE) is proposed. Firstly, the correlation between genes and class tags is evaluated by Su, and the Markov blanket is approximated according to the definition of Su. A large number of unrelated and redundant genes are eliminated quickly. Then the redundant genes are further eliminated by SVM-RFE, and an effective subset of information genes is obtained. The proposed method can obtain higher tumor classification performance under the condition of selecting a subset of information genes with less dimension or equal dimension.
【作者单位】：皖南医学院计算机教研室;安徽师范大学数学计算机科学学院;
【基金】：国家自然科学基金项目(No.61672386) 教育部人文社会科学研究规划基金项目(No.16YJAZH071) 安徽省自然科学基金项目(No.1708085MF142) 安徽高校省级自然科学研究重点基金项目(No.KJ2016A275,KJ2014A266)资助~~
【分类号】：R73;TP18

【参考文献】