大样本线性回归模型的子抽样及变量选择  

Subsampling and Variable Selection of Linear Regression Model With Large Samples

作　　者：徐礼文 廖丹 Xu Liwen;Liao Dan(College of Science,North China University of Technology,Beijing 100144,China)

机构地区：[1]北方工业大学理学院,北京100144

出　　处：《统计与决策》2022年第2期5-9,共5页Statistics & Decision

基　　金：国家社会科学基金资助项目(20BTJ046)。

摘　　要：文章考虑了大样本下线性回归中同时进行快速估计和变量选择的问题,即针对一个存在稀疏解的大样本线性模型,根据重要性抽样分布从全数据集抽取少量子样本,对该子样本进行自适应Lasso估计。通过随机模拟研究,将该算法分别应用在几种不同的数据集中,并从模型预测精度和可解释性两个方面比较了四种子抽样方法在该算法下的表现。模拟结果表明,所提出的算法具有良好表现,在计算开销上也具有一定优势。This paper considers the problem of fast estimation and variable selection in linear regression with large samples,that is,for a large sample linear model with sparse solutions,the sampling distribution extracts a small number of subsamples from the whole data set according to the importance,and estimates the sub samples by adaptive Lasso.The algorithm is applied to several different data sets by random simulation,and the performance of four subsampling methods is compared in terms of prediction accuracy and interpretability of the model.Simulation results show that the proposed algorithm performs well and has some advantages in computational overhead.

关 键 词：大样本 数据降维 子抽样算法 变量选择 

分 类 号：O212.1[理学—概率论与数理统计]