基于条件信息熵的超高维分类数据特征筛选  

Feature Screening in Ultrahigh Dimensional Categorical Data Based on Conditional Information Entropy

作　　者：程国胜[1] 孙超男 宋凤丽[1] 来鹏 Cheng Guosheng;Sun Chaonan;Song Fengli;Lai Peng(School of Mathematics and Statistics, Nanjing University of Information Science and Technology, Nanjing 210044, China)

机构地区：[1]南京信息工程大学数学与统计学院

出　　处：《统计与决策》2018年第8期64-67,共4页Statistics & Decision

基　　金：国家自然科学基金资助项目(11301279);江苏省自然科学基金资助项目(BK20140983;BK20161530);江苏省“青蓝工程”资助项目(2016)

摘　　要：文章提出了一种基于条件信息熵的超高维自由模型下非参数特征筛选方法,在响应变量为两类别时,对多类别离散型协变量进行特征筛选。通过理论证明和蒙特卡罗数值模拟验证了该筛选方法具有确定筛选性质,对超高维分类变量的重要特征筛选具有显著的效果。This paper proposes a new nonparametric free-model screening method based on the conditional information entropy for ultrahigh dimensional data analysis, and performs features screening of many types of discrete covariates. And finally a theoretical verification and a Monte Carlo simulation have been made to demonstrate that the proposed screening technique possesses a confirmed screening property and is obviously effective for screening the significant features of ultra-high dimensional categorical variable.

关 键 词：超高维判别分析 特征筛选 确定筛选性质 条件信息熵 

分 类 号：O212.4[理学—概率论与数理统计]