Logistic回归模型中参数极大似然估计的二次下界算法及其应用

QUADRATIC LOWER-BOUND ALGORITHM FOR MAXIMUM LIKELIHOOD ESTIMATOR OF LOGISTIC REGRESSION ON PARAMETER AND ITS APPLICATION

本文研究了Newton-Raphson等算法无法进行时探寻更加稳定的数值解法的问题.利用B¨ohning&Linday(1988)提出的二次下界算法(Quadratic lower-bound),文中在Logistic回归模型下构造了极大似然函数的代理函数并进行数值模拟,获得了二次下界算法是Newton-Raphson算法的合理替代的结果,推广了数值方法在Logistic回归模型中的应用.

In this paper,we study how to explore more stable numerical solution when parameters cannot be solved by using Newton-Raphson algorithm.By using the quadratic lower bound algorithm that B¨ohning Linday has proposed in 1988,we construct a surrogate function for maximum likelihood function under Logistic regression model and the simulation results verify that quadratic lower bound algorithm is a reasonable algorithm of Newton-Raphson algorithm,which extend numerical method＇s application under Logistic regression model.