超高维可加模型下的特征筛选
首发时间:2015-04-23
摘要:Fan and Lv(2008) 提出了以相关系数为基础的安全独立筛选方法,简称为SIS,此种方法可以对超高维模型进行特征筛选。然而,此种方法并不适用于所有模型。当边际回归为非线性时,此种方法筛选出的结果没有安全筛选性质。本篇文章提出了以核函数为基础,并适用于超高维可加模型的特征筛选方法 。在一定条件下,此种方法拥有安全筛选性质。模拟结果表明,在一定样本数和较大维数条件下,本文所提出的方法拥有良好的筛选结果。
For information in English, please click here
Screening for Sparse Ultra High Dimensional Additive Models
Abstract:Fan and Lv(2008) developed a variable screening procedure based on correlation learning, named sure independence screening (SIS), which can reduce dimensionality in ultra-high-dimensional models. However, this method is not suitable for all models. When the marginal regression is not linear, the method of SIS in Fan and Lv(2008) does not have sure screening property. In this paper we propose feature screening method in the ultra high dimensional additive models based on kernel method. Under some mild technical conditions, the sure screening property is valid for the proposed method. The simulation results indicate that the proposed procedure works well with moderate sample size and large dimension.
Keywords: Feature screening Ultra-high-dimensional additive models Local linear aproximate.
基金:
论文图表:
引用
No.4639218105278414****
同行评议
共计0人参与
勘误表
超高维可加模型下的特征筛选
评论
全部评论0/1000