基于核函数的混合C均值聚类算法

提出了一种基于核函数的混合C均值聚类算法.首先利用模糊C均值聚类算法和另一种类型的可能性C均值聚类算法的优点,设计出一种混合C均值聚类算法.然而鉴于该算法存在的不足,本文将Mercer核函数引入到该算法中,仿真实验结果证实了该方法的可行性和有效性.     

缺失数据下联合均值与方差模型的参数估计

基于正态分布提出了缺失数据下联合均值与方差模型,在响应变量随机缺失下研究了该模型均值插补、回归插补和随机回归插补三种插补方法的参数估计,通过数据模拟和实例研究结果比较表明,随机回归插补方法是三种插补方法中最有用和有效的。     

一种稳健的聚类方法

本文讨论一种新的聚类方法 :属性均值聚类 .通过理论分析 ,属性均值聚类是比模糊均值聚类更稳健的聚类方法 .数值实验说明了该方法的有效性     

加权模糊C-均值聚类

将经典的模糊C-均值聚类中的欧氏距离推广到广义欧氏距离,得到了加权模糊C-均值聚类的迭代公式,实证分析表明加权模糊C-均值聚类的结果与主成分分析的排序基本一致,特别适用于大样本的聚类与排序。     

NMAR机制下非线性均值方差模型的Bayes估计

讨论响应变量带有不可忽略缺失数据的非线性均值方差模型的Bayes估计问题.缺失数据机制由logistic回归模型来指定,运用Gibbs抽样及MH算法得到模型参数和缺失数据机制参数的联合Bayes估计,模拟研究和实例分析展示上述模型和方法的可行性.     

一种基于核的混合c-均值聚类模型

FCM和PCM的混合模型可以克服它们单独聚类时的缺点,在聚类效果上有很大改进,但是对于特征不明显的样本而言,这种混合模型的聚类效果并不太好,为了克服这一缺点,本文引入Mercer核,提出了一种新的基于核的混合c-均值聚类模型(KIPCM),运用核函数使得在原始空间不可分的数据点在核空间变得可分。通过数值实验,得到了较为合理的中心值以及较高的正确分类率,证实了本文算法的可行性和有效性。     

自适应约束模糊C均值聚类算法

针对经典C均值聚类算法和模糊C均值聚类算法所存在的对初始聚类中心过分依赖以及需要预先知道实际聚类数目的问题,基于模糊C均值聚类算法提出了一种新算法:自适应约束模糊C均值(ACFCM)聚类算法,它在模糊C均值聚类算法的基础上,给目标函数加入了一个惩罚项,使得上述问题得以解决.并通过仿真实验证实了新算法的可行性和有效性.     

均值及均值方程

均值及均值方程是函数方程领域近二十年研究的热门方向.本文首先介绍了均值的定义及性质、Gauss迭代、不变方程和M-S问题的发展历程.接着围绕Bajraktarevi?均值、Cauchy均值及其他相关均值进行综述,尤其详述了对应的相等性与不变性问题的研究进展.Bajraktarevi?均值和Cauchy均值这两类均值均含两个衍生函数,对应的等式问题已基本得以解决,但这两类对称均值的不变性因含有四个未知函数而求解困难的问题目前仍未完全解决.最后介绍了均值在其他领域中的应用.     

二分类Logistic回归插补法及其应用

在实际的调查数据和实验数据中,经常会出现数据缺失的问题,插补方法是处理缺失数据的一种常用的技术方法.对于目标变量是二分类的定性变量时,可以采用Logistic回归插补法进行插补,采用一套高中生进入大学学习影响因素分析的模拟数据进行实证分析,探讨了Logi8tic回归插补法的一些特点.     

κ-均值聚类算法的改进及其在冰脊表面形态分析中的应用

针对传统k-均值聚类算法事先必须获知类别数和难以确定初始聚类中心的缺点,建立了关于聚类中心和类别数k的双层规划模型,结合粒子群算法确定出聚类中心,通过在迭代过程中不断更新准则函数的方法搜索并确定出最佳类别数惫,基于所建模型,提出了一种改进的k-均值聚类算法,并将算法应用于冰脊表面形态分析中.结果表明,算法得到的聚类结果不但具有相邻类别边界清晰的优点,而且能够较好地反映出地理位置和生长环境对冰脊形成的影响.     

Imputation of mean of ratios for missing data and its application to PPSWR sampling

In practical survey sampling, nonresponse phenomenon is unavoidable. How to impute missing data is an important problem. There are several imputation methods in the literature. In this paper, the imputation method of the mean of ratios for missing data under uniform response is applied to the estimation of a finite population mean when the PPSWR sampling is used. The imputed estimator is valid under the corresponding response mechanism regardless of the model as well as under the ratio model regardless of the response mechanism. The approximately unbiased jackknife variance estimator is also presented. All of these results are extended to the case of non-uniform response. Simulation studies show the good performance of the proposed estimators.     

Sample rotation theory with missing data

This paper studies how the sample rotation method is applied to the case where item non-response occurs in surveys. The two cases where the response to the first occasion is complete or incomplete are considered. Using ratio imputation method, the estimators of the current population mean are proposed, which are valid under uniform response regardless of the model and under the ratio model regardless of the response mechanism. Under uniform response, the variances of the proposed estimators are derived. Interestingly, although their expressions are similar, the estimator for the case of incomplete response on the first occasion can have smaller variance than the one for the case of complete response on the first occasion under uniform response. The linearized jackknife variance estimators are also given. These variance estimators prove to be approximately design-unbiased under uniform response. It should be noted that similar property on variance estimators has not been discussed in literature.     

缺失数据下广义变系数模型的均值借补估计

本文在响应变量随机缺失时,给出广义变系数模型中响应变量的2个均值拟似然借补估计。证明了它们具有渐近正态性,并进行了模拟研究。     

带有缺失数据的广义半参数模型的均值借补估计

本文在响应变量随机缺失时, 给出了广义半参数模型中响应变量的2个均值拟似然借补估计.证明了它们具有渐近正态性, 给出了估计的渐近偏差与渐近方差, 并进行模拟比较.     

抽样调查中缺失数据的插补方法

在抽样调查等实际问题中,经常出现数据缺失.针对这类问题,通常的处理方法之一是对数据进行插补。本文综述了抽样调查中处理缺失数据常用的插补方法。重点讨论了单一插补的方差估计与多重插补的简化计算以及使用回答概率的单一插补等。最后讨论目前插补所面临的问题与其发展方向.     

���ڳ�����α�������������ĺ�ѡ�����ݿ����������ƶ��о�

??How to solve the inference problem of candidate database web surveys is an urgent problem to be solved in the development of web survey. In order to solve this problem, the inference method of non-probability sampling based on superpopulation pseudo design and the combined sample is proposed. A superpopulation model is firstly built up to construct pseudo weights for a survey sample of the web candidate database. The estimator of the population mean is then computed according to the combined sample composed of the survey sample of the web candidate database and a probability sample. The variance estimator of the population mean estimator is lastly derived according to the variance estimation theory of the superpopulation model. The Bootstrap and Jackknife methods are also used to compute the variance estimator. And all these variance estimation methods are compared. The research results show that the population mean estimator based on superpopulation pseudo design and the combined sample is better, and has higher efficiency than the estimator only using the probability sample and the weighted estimator only using the survey sample of the web candidate database. The variance estimator computed by using the VM1, VM2 and VM3 method are relatively better.     

Research on Inference of Candidate Database Web Surveys Based on Superpopulation Pseudo Design and the Combined Sample

How to solve the inference problem of candidate database web surveys is an urgent problem to be solved in the development of web survey. In order to solve this problem, the inference method of non-probability sampling based on superpopulation pseudo design and the combined sample is proposed. A superpopulation model is firstly built up to construct pseudo weights for a survey sample of the web candidate database. The estimator of the population mean is then computed according to the combined sample composed of the survey sample of the web candidate database and a probability sample. The variance estimator of the population mean estimator is lastly derived according to the variance estimation theory of the superpopulation model. The Bootstrap and Jackknife methods are also used to compute the variance estimator. And all these variance estimation methods are compared. The research results show that the population mean estimator based on superpopulation pseudo design and the combined sample is better, and has higher efficiency than the estimator only using the probability sample and the weighted estimator only using the survey sample of the web candidate database. The variance estimator computed by using the VM1, VM2 and VM3 method are relatively better.     

New imputation methods for missing data using quantiles

The problem of missing values commonly arises in data sets, and imputation is usually employed to compensate for non-response. We propose a novel imputation method based on quantiles, which can be implemented with or without the presence of auxiliary information. The proposed method is extended to unequal sampling designs and non-uniform response mechanisms. Iterative algorithms to compute the proposed imputation methods are presented. Monte Carlo simulations are conducted to assess the performance of the proposed imputation methods with respect to alternative imputation methods. Simulation results indicate that the proposed methods perform competitively in terms of relative bias and relative root mean square error.     

响应变量随机缺失下的半参变系数模型的均值估计

在响应变量随机缺失时,研究了半参数变系数模型响应变量均值的借补估计.首先利用完整个体估计模型中的参数与非参数部分,然后再用借补方法与加权借补方法估计响应变量的均值.最后求出了估计的渐近偏差与渐近方差,研究了所得到的估计的渐近性质,并进行模拟比较.     

基于收缩近邻方法的征信缺失数据插补研究

在海量征信数据的背景下,为降低缺失数据插补的计算成本,提出收缩近邻插补方法.收缩近邻方法通过三阶段完成数据插补,第一阶段基于样本和变量的缺失比例计算入样概率,通过不等概抽样完成数据的收缩,第二阶段基于样本间距离,选取与缺失样本近邻的样本组成训练集,第三阶段建立随机森林模型进行迭代插补.利用Australian数据集和中国各银行数据集进行模拟研究,结果表明在确保一定插补精度的情况下,收缩近邻方法较大程度减少了计算量.