摘要
中位数回归不对误差项分布做过强假设且对异常值不敏感,可以提高回归模型的稳健性。自适应LASSO进行变量选择时对自变量采用有差别的惩罚系数,避免了系数的过度压缩。对于含有顺序类别自变量数据进行回归建模时,考虑到此类自变量中伪分类的存在。构建了一种通过哑变量的线性变换,并结合自适应LASSO惩罚的中位数回归方法。该方法不仅能够进行变量选择得到稳健的估计结果还能进行伪分类的识别与融合。通过2个实际数据验证了该方法的可行性和有效性。
Median regression does not make strong assumptions about error term distribution and is insensitive to outliers,which can improve the robustness of the regression model.When selecting variables,adaptive LASSO adopts differential penalty coefficients for independent variables to avoid excessive compression of coefficients.The pseudo classification of such covariates is considered in the regression modeling which contains ordinal categorical covariates.Combined with a penalized median regression method for adaptive LASSO,a linear transformation through dummy variables is constructed.This method can not only get robust estimation results through variable selection but also identify and fuse pseudo classification.Finally,the feasibility and effectiveness of the proposed method are demonstrated by two real data.
作者
吉洋莹
潘雨辰
黄磊
JI Yangying;PAN Yuchen;HUANG Lei(School of Mathematics,Southwest Jiaotong University,Chengdu 611756,China)
出处
《重庆理工大学学报（自然科学）》
CAS
北大核心
2022年第11期257-265,共9页
Journal of Chongqing University of Technology：Natural Science
基金
国家自然科学基金项目(11771066)
中央高校基本科研业务费专项资金项目(2682020ZT113)
四川省自然科学基金项目(2022NSFSC1850)。
关键词
自适应LASSO
中位数惩罚回归
顺序类别自变量
伪分类
adaptive LASSO
penalized median regression
ordinal categorical covariates
pseudo classification
