“碎片化数据”的模型平均方法

发布时间：2018-01-19 06:25

本文关键词： 碎片化数据线性模型异方差刀切法模型平均最优权重　出处：《华东师范大学》2016年硕士论文　论文类型：学位论文

【摘要】：随着“大数据”时代的来临,越来越多的数据获得渠道给科学研究及其实际应用带来了巨大的机遇。但另一方面,由于每个样本可以获得的数据渠道不尽相同,我们面临的可能是一种“碎片化”的数据,这给统计建模带来了很大的挑战。在这篇文章里,我们提出一种基于模型平均的方法来处理“碎片化数据”。不同于传统统计方法依赖于单一的模型,模型平均方法可以根据数据的可获得情况拟合多个模型,然后通过加权平均来获得最终的估计。我们在可获得完整数据的那部分样本上采用刀切法来获得加权的权重并严格证明了该权重选择方法的最优性。我们做了大量的数值模拟来验证所提出的方法在有限样本上的优异表现。同时我们通过一个实例分析展现了我们方法的实际应用。
[Abstract]:With the advent of "big data" era, more and more channels of data acquisition have brought great opportunities for scientific research and its practical application. On the other hand, because of the different data channels each sample can obtain. We may face a "fragmentation" of data, which poses a great challenge to statistical modeling, in this article. We propose a method based on model averaging to deal with "fragmentation data". Different from the traditional statistical method which relies on a single model, the model averaging method can fit multiple models according to the availability of data. Then we get the final estimate by weighted average. We use the knife cutting method to obtain the weighted weight in the sample where the complete data can be obtained, and prove strictly the optimality of the weight selection method. The numerical simulation of the proposed method is used to verify the performance of the proposed method on a finite sample. At the same time, we demonstrate the practical application of our method by an example analysis.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：C81

【相似文献】