期刊文献+

一种基于近似马尔可夫毯的mRMR-PCA降维方法

MRMR-PCA Dimension Reduction Method Based on Approximate Markov Blanket
下载PDF
导出
摘要 机器学习过程中样本数据含有过多冗余或不相关特征时,会增加学习模型的复杂度,降低模型学习的准确率。针对样本冗余或不相关特征删除的问题,提出了基于近似马尔可夫毯的最大相关最小冗余法与主成分分析法的二阶段特征降维方法。改进算法结合马尔可夫毯和最小冗余最大相关算法的优势,以互信息为度量标准,根据特征与目标类的相关度、特征与特征的冗余筛选样本特征子集,并利用主成分分析法对所筛选的特征子集进行压缩,形成用于分类的综合特征集。最后以SVM、BP和KNN作为分类器进行样本分类实验,实验结果表明,改进算法在特征降维、分类准确率和分类耗时上均优于MB-mRMR算法与PCA算法。 The sample data in the maehinr learning conlains too many redundant or irrelevanl features,which increases the complexity of the learning model and reduces the accuracy of model learning.In order to delete redundant or inelevant features in samples,a two-stagr feature dimension rrduclion melhod is proposed,which is a combination of the Minimum Redundancy Maximum Relevance(mRMR)based oil approximate Markov blanket and the Principal Component Analysis(PCA).Mutual infonnation is used as the metric of the features selection.Firstly the mulual information of the feature and the target class,feature and feature is calcukited,then the feature subsets of sample features are fillered by approximate Markov Blanket and mRMR algorithm.Secondly,in order to further reduce the dimension of features.PCA algorithm is used to filter feature subsets to form a comprehensive feature set,which is ullimately used for classification.In the simulation experiments.Support vector Machine(SVM),Back Propagation(BP)and k-NearestNeighbor(KNN)are used as classifiers to carry out classification.The experimental results show that the improved algorithm is superior to MB-mRMR and PcA in feature dimension reduction,classification accuracy and classification time-consuming.
作者 徐明月 邱均平 林泽轩 顾彦 Xu Mingyue;Qiu Junping;Lin Zexuan;Gu Yan(School of Computer,Hangzhou Dianzi University,Hangzhou 310018;Chinese Academy of Science and Education Evaluation,Hangzhou Dianzi University,Hangzhou 310018)
出处 《评价与管理》 2022年第1期50-55,共6页 Evaluation & Management
关键词 互信息 近似马尔可夫毯 mRMR算法 PCA 降维 Mutual information Approximate Markov blanket mRMR PCA Dimensionality reduction
  • 相关文献

参考文献2

二级参考文献15

共引文献7

相关作者

内容加载中请稍等...

相关机构

内容加载中请稍等...

相关主题

内容加载中请稍等...

浏览历史

内容加载中请稍等...
;
使用帮助 返回顶部