基于Lasso特征选择的方法比.pdfVIP

年第 期 安徽电子信息职业技术学院学报 2014 1 No.1 2014 第 卷 总第 期 13 ( 70 ) JOURNALOF ANHUI VOCATIONAL COLLEGE OF ELECTRONICS ＆ INFORMATION TECHNOLOGY General No.70 Vol.13 [文章编号] 1671－802X(2014)01－0026－05 基于 ????? 特征选择的方法比较 刘晓宁 （太原科技大学， 山西 太原 030041） ［摘 要］模型和特征选择是统计学中较为重要的问题之一。Lasso 是一种基于一范式的特征选择方法， 与现有特征选择方法比较，Lasso 不仅能够准确地选择出重要变量，同时还具有特征选择的稳定性。文中对线 性回归模型中变量选择的 Lasso 算法、基于线性模型的 Lasso、Lars???Adaptive-lasso、elastic net 等方法进行了 比较，指出了它们间的联系，并通过对几个选自 UCI 数据集的数据进行对比验证，给出了变量选择方法的具 体实现。 ［关键词］特征选择；Lasso 算法；线性回归；变量选择 ［中图分类号］O213.9 ［文献标识码］A Comparison of Feature Selection Methods Based on Lasso LIU Xiao-ning (Taiyuan University of Science and Technology, Taiyuan 030041, Shanxi) Abstract: The model and feature selection is one of the important subjects in statistics. Lasso is a feature selection method based on 1-norm. Compared with the existing features of selection methods, Lasso can not only accurately choose the important variables, but also has the stability of feature selection. This paper compares the Lasso algorithm of variable selection in linear regression model, and Lasso, Lars, the Adaptive Lasso, elastic net as well as other methods which are based on the linear model. The relationships between them are presented. A variable selection method is realized by the comparison tests of a few data selected from UCI. Key words: feature selection; Lasso algorithm; linear regression; variable selection 现代统计学的迅速发展与高速电子计算机的不 小二乘达不到预测的准确性还有不能有较好的解释