高维数据在Cox回归模型中的自变量选择——基于Elastic Net方法的维数约简

Independent Variable Selection of High-Dimensional Data in Cox Regression Model—Dimension Reduction Based on Elastic Net

在高维数据分析中,LASSO维数约简方法占有很重要的位置。针对数据愈发繁杂,LASSO回归不再适应一些具有较高相关性的高维数据分析,由此产生了Elastic Net和其他相关的一些高维数据拓展分析方法。Elastic Net是在LASSO的思想方法基础上结合非凸罚函数和岭回归方法得到的,Adaptive Elastic Net等是在Elastic Net的思想方法上,通过数据特征的不同改进惩罚函数,缓和稀疏性和过拟合的问题。文章对Elastic Net、Adaptive Elastic Net、Weight Elastic Net进行了介绍并且通过实际例子做了简单的比较,最终得到了较好的维数约简方法。

In high-dimensional data analysis, LASSO occupies a very important position. For data becoming more and more complicated, LASSO regression is not very suitable for some relevant high-dimen- sional data analysis. Some of the extended analysis methods are produced, such as adaptive Elastic Net and other high-dimensional data analysis methods. Elastic Net is obtained by combining non- convex penalty and ridge regression methods on the basis of LASSO’s method of thinking. Adaptive Elastic Net, etc., based on Elastic Net’s method of thinking, uses different data characteristics to improve the penalty function, and continuously corrects the sparsity and overfitting problem. The article mainly introduces Elastic Net, Adaptive Elastic Net, Weight Elastic Net and makes a simple comparison of several methods through practical examples.