基于输出不一致测度的极限学习机集成的基因表达数据分类  

Disagreement Measure Based Ensemble of Extreme Learning Machine for Gene Expression Data Classification

作　　者：陆慧娟[1,2] 安春霖[2] 马小平[1] 郑恩辉[3] 杨小兵[2] 

机构地区：[1]中国矿业大学信息与电气工程学院,江苏徐州221008 [2]中国计量学院信息工程学院,杭州310018 [3]中国计量学院机电工程学院,杭州310018

出　　处：《计算机学报》2013年第2期341-348,共8页Chinese Journal of Computers

基　　金：国家自然科学基金(61272315;60842009;60905034);浙江省自然科学基金(Y1110342;Y1080950);浙江大学包氏留学基金资助~~

摘　　要：选择性集成学习已经成为分析基因表达数据、获取生物学信息的有力工具.为了更好地挖掘基因表达数据,利用极限学习机的集成,克服单个ELM用于数据分类时性能欠稳定的缺点,文中提出了一种基于输出不一致测度的ELM相异性集成算法(D-D-ELM).算法首先以输出不一致测度为标准对多个ELM模型进行相异性判断,其次根据ELM的平均分类精度剔除掉相应的模型,最后对筛选后的分类模型用多数投票法进行集成.算法被运用到Breast、Leukemia、Colon、Heart基因表达数据上,并通过理论和实验得到验证.实验结果的统计学分析表明D-D-ELM能够以更少的模型数量达到较稳定的分类精度.Selective ensemble learning has become a powerful tool for biological information anal- ysis of gene expression data. In order to mining gene expression data better, we use ensemble of Extreme Learning Machine （ELM） to overcome the shortage that a single ELM is unstable in data classification. In this paper, we propose an algorithm, which is the dissimilarity ensemble based on disagreement measure of Extreme Learning Machines （D-D-ELM）. First, we judge the dis- similarity of Extreme Learning Machines with disagreement measure. Then we remove the corre- sponding ELMs based on the average classification accuracy. At last, the rest ELMs are grouped into an ensemble classifier by the strategy of majority voting. This algorithm is applied on the data of gene expression Breast, Leukemia, Colon, Heart. The theoretical analysis and experi- ment are given and the statistical analysis on the experimental results demonstrates that D-D-ELM can achieve better classification accuracy with less number of ELMs.

关 键 词：极限学习机 基因表达数据 集成算法 输出不一致测度 分类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]