摘要
大数据时代,高维数据的变量选择是现代统计的研究热点问题之一。MCP正则化方法是常用的变量选取方法,但MCP正则化方法的优劣取决于能否选取出最优的正则化参数。本文在BIC准则的基础上,提出适用于MCP正则化参数选择的MBIC准则。通过数据模拟及实际应用表明,MCP方法在MBIC准则下能够以更高的概率选择正确的模型,MBIC准则明显优于其它参数选择方法。
In the era of big data, variable selection of high-dimensional data is one of the hot topics in modern statistics. The MCP regularization method is a commonly used variable selection method, but the merits of the MCP regularization method depend on whether the optimal regularization parameter can be selected. Based on the BIC criterion of regularization parameter selection, an MBIC criterion is proposed for MCP regularization parameter selection. Through data simulation and practical application, the MCP method with MBIC criterion can select the correct model with higher probability, which is obviously superior to other regularization parameter selection methods.
出处
《统计学与应用》
2019年第6期852-858,共7页
Statistical and Application
基金
浙江省自然科学基金资助项目(LY18A010026)
全国统计科学研究项目(2019LY06)
宁波市自然科学基金资助项目(2017A610143)
国家级大学生创新创业训练计划项目(201911058025)
王伟明助创基金资助项目(2018020)
浙江省大学生科技创新活动计划暨新苗人才计划资助项目(2018R428027)。
