rLasso正则化Logistic回归模型的估计

Estimation of Logistic Regression Model Based on rLasso Regularization

文章将rLasso惩罚函数推广到Logistic回归模型,并给出单坐标rLasso惩罚估计问题的解析解,结合坐标下降算法思想,给出线性模型rLasso以及Logistic-rLasso惩罚估计问题的坐标下降求解方法。数值模拟验证所提坐标下降算法的有效性,并说明rLasso惩罚比LASSO类惩罚能选择更为稀疏的模型。

This paper extends rLasso penalty function to Logistic regression model, and proposes an analytical solution to a single coordinate rLasso. The paper also combines with coordinate descent algorithm to give the coordinate descent method of lin ear model rLasso and Logistic-rLasso penalty estimation. Numerical simulations verify the effectiveness of the proposed coordi- nate descent algorithm and indicates that rLasso penalty can choose a more sparse model than Lasso penalty.