含多元非线性结构的高维Logistic模型的同时变量选择及半参数估计-概率论与数理统计专业论文.docxVIP

摘要摘要 摘要 摘要 众所周知，Logistic模型是一种广义线性回归分析模型，在医药学、经济学 等领域都有重要的应用。本文将介绍具有多元非线性结构的高维Logistic回归 模型，其线性部分是高维的，非线性部分为多元函数。文中采用Lasso方法，并 结合EBIC准则对模型的线性部分进行变量选择及参数估计：对非线性部分的 处理分为两种：估计和变量选择。本文将利用RKHS(再生核希尔伯特空间)理 论，借鉴SS—ANOVA方法，并结合GCV准则对非线性部分进行估计；我们同 样的采用Lasso方法并结合EBIC准则对非线性部分进行变量选择。在方法实现 上，线性部分和非线性部分的处理过程中均有控制参数。文中还给出了大量的 模拟结果，进而分析和比较本文所用算法的可行性及优劣性。最后通过分析一 个实际例子来说明文章所用方法的实用性以及优越性。本文的主要工作为：提 出适用于本文模型的算法，并进行大量的数值模拟及实证分析。 关键词： 高维，含多元非线性结构的Logistic模型，Lasso，RKHS，张量积 RKHS，SS．ANOVA，EBIC，GCV 万方数据 ABSTRACTABSTRACT ABSTRACT ABSTRACT As everyone knows，the Logistic model is generalized linear regression analysis model．It has important applications in medicine．economics and other fields．’nlis paper will introduce the hiIgh dimensional Logistic regression model with multivariate nonlinear structure．The model has high dimensional linear part and multiple nonlinear part．In this paper,we perform variable selection and parameter estimation ofthe linear part by using the method of Lasso combined with EBIC criterion．The processing ofthe nonlinear part is divided into two types：estimation and variable selection．This paper will use the theories of RKHS(reproducing kernel Hilbert space)and the methods based on the SS·ANOVA，then combined with GCV criterion to estimate the nonlinear part． We also used the Lasso method combined with EBIC criterion for variable selection of the nonlinear part．In the method，the realization process of linear and nonlinear parts all have tuning parameters．A large number of simulation results are also given in this paper,and then the feasibility and advantages of the algorithm are analyzed and compared．Finally,an example is used to demonstrate the pmcticability and superiority ofthe method used in this paper．The main work ofthis paper is：the introduction ofthe algorithm for the model，a large number of numerical simulation and empirical analysis of the algorithm． Keywords：High-dimensional，Logistic regression model with multivariate nonlinear structure，Lasso，RKHS，Tensor