代价敏感特征选择的肿瘤分类应用  

Application of cost-sensitive feature selection algorithm in tumor gene expression data classification

作　　者：边婧[1,2] 彭新光[1] 张海 BIAN Jing PENG Xin-guang ZHANG Hai(College of Computer Science and Technology, Taiyuan University of Technology, Taiyuan 030024,China Center of Information and Network, Shanxi Medical College of Continuing Education, Taiyuan 030012, China Department of Technology and Product Management, Shanxi Branch of Agricultural Bank of China, Taiyuan 030024, China)

机构地区：[1]太原理工大学计算机科学与技术学院,山西太原030024 [2]山西职工医学院网络信息中心,山西太原030012 [3]中国农业银行山西省分行科技与产品管理部,山西太原030024

出　　处：《计算机工程与设计》2017年第5期1342-1346,1388,共6页Computer Engineering and Design

基　　金：山西省卫生厅科研基金项目(201301006);山西省研究生优秀创新基金项目(20123030)

摘　　要：针对肿瘤基因数据高维、小样本的特点,引入误分类代价,基于混沌遗传算法提出一种代价敏感特征选择算法。结合误分类代价矩阵与分类精度,构建基于最近邻的代价敏感特征选择适应度函数,在降低特征选择总成本的同时权衡代价成本与分类精度,采用混沌遗传算法提高搜索效率,提升算法性能。将该方法应用于肿瘤基因数据进行有效性验证,实验结果表明,该算法降低了特征空间维数,有效提高了肿瘤的分类性能。The tumor gene expression data is usually very high dimensional with small sample. Aiming at the characteristics of gene, an efficient cost-sensitive feature selection based on chaos genetic algorithm was proposed. By combining with two key fac-tors of average total misclassification cost and classification accuracy, the cost-sensitive fitness function based on the most neigh-bor classifying function was constructed. The total cost of feature selection was reduced and both factors were balanced. Chaos genetic algorithm was applied to improve search efficiency and algorithm performance. Application results in tumor gene expres-sion data show that the proposed algorithm effectively and efficiently reduces the feature dimensions and improves the classifica-tion accuracy.

关 键 词：代价敏学习 特征选择 肿瘤基因数据 混沌遗传算法 分类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]