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

作　　者：王佳[1] 丁洁丽[1] 

机构地区：[1]武汉大学数学与统计学院,湖北武汉430072

出　　处：《数学杂志》2015年第6期1521-1532,共12页Journal of Mathematics

基　　金：国家自然科学基金(11101314)

摘　　要：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.

关 键 词：minorization-maximization算法 LOGISTIC回归模型 QUADRATIC lower-bound算法 极大似然估计 Newton-Raphson算法 

分 类 号：O212.1[理学—概率论与数理统计]