基于互信息的变量选择方法

Variable Selection Method Based on Mutual Information

文章基于解释变量与被解释变量之间的互信息提出一种新的变量选择方法:MI-SIS。该方法可以处理解释变量数目p远大于观测样本量n的超高维问题,即p=O(exp(nε))ε>0。另外,该方法是一种不依赖于模型假设的变量选择方法。数值模拟和实证研究表明,MI-SIS方法在小样本情形下能够有效地发现微弱信号。

This paper proposes a new variable screening method based on mutual information between explanatory variables and explained variables,namely MI-SIS.This method can deal with the ultrahigh dimension problem where the number of explanatory variables[p]is much larger than the observed sample size[n],that is,[p=O(exp(nε)),ε>0.].In addition,the proposed method is a variable selection method independent of model assumptions.Numerical simulation and empirical study show that the MI-SIS method can effectively detect weak signals in small samples.