您的位置:首页  > 论文页面

Lasso与其他变量选择方法的模拟比较

发表时间:2011-01-15  浏览量:4311  下载量:1888
全部作者: 胡一睿,曲荣华,陈思扬,徐佳静,童行伟
作者单位: 北京师范大学数学科学学院;北京师范大学经济与工商管理学院
摘 要: 目的: 引入一种基于收缩估计的新变量选择方法——Lasso (the least absolute shrinkage and selectionator),并比较其与其他变量选择方法的异同。方法: 首先给出了几种常见的变量选择方法,如逐步回归、赤池信息准则(Akaike information criteria, AIC)、贝叶斯信息准则(Bayesian information criterions, BIC),再通过随机模拟给出了几种方法进行变量选择的结果及相关准确性分析。结果: 随机模拟结果表明,当模拟次数n=200时,Lasso方法的平均错误率已经为0,具有较为明显的优势,随着模拟次数的增加,Lasso方法的平均正确率(0.951)达到了相对较高的水平。结论: Lasso估计具有较好的可解释性,在变量选择中有较广阔的应用前景。
关 键 词: 理论统计学;变量选择; Lasso;赤池信息准则;贝叶斯信息准则;逐步回归
Title: Simulation and comparison of Lasso and other variable selection methods
Author: HU Yirui, QU Ronghua, CHEN Siyang, XU Jiajing, TONG Xingwei
Organization: School of Mathematical Sciences, Beijing Normal University; School of Economy and Business Administration, Beijing Normal University
Abstract: Objective: To bring in a new variable selection method: Lasso (the least absolute shrinkage and selectionator) based on shrinkage estimate, and then compare it with other variable selection methods. Methods: First, analyzing several common variable selection methods such as stepwise regression, Akaike information criterion (AIC) and Bayesian information criterion (BIC). After stochastic simulation, comparing correct and incorrect rates of those methods. Results: The stochastic simulation showed that Lasso’s incorrect rate had decreased to 0 when n=200, with obvious advantages. As the simulation times increased, the average correct rate of Lasso (0.951) had reached a relatively higher level. Conclusion: Lasso method is easy to understand, and it has broad application prospects in variable selection.
Key words: theoretical statistics; variable selection; Lasso; Akaike information criteria; Bayesian information criterion; stepwise selection
发表期数: 2011年1月第1期
引用格式: 胡一睿,曲荣华,陈思扬,等. Lasso与其他变量选择方法的模拟比较[J]. 中国科技论文在线精品论文,2011,4(1):63-66.
 
0 评论数 0
暂无评论
友情链接