半参数方法与BP神经网络在缺失数据中的对比研究  

Comparation of Semiparametric Method and BP Neural Network in Missing Data

作　　者：翟芳慧 施三支[1] 樊思敏 ZHAI Fang-hui;SHI San-zhi;FAN Si-min(School of Science,Changchun University of Science and Technology,Changchun 130022)

机构地区：[1]长春理工大学理学院,长春130022

出　　处：《长春理工大学学报（自然科学版）》2020年第3期115-120,共6页Journal of Changchun University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金(11601039);吉林省自然科学基金(20140101199JC)。

摘　　要：数据缺失的存在不仅会增大统计分析的复杂性和难度,还会导致分析结果的偏倚。比较了半参数方法和BP神经网络方法在分量指标数据符合正态分布且为随机缺失机制情况下的优劣。选取鸢尾花数据集进行模拟研究,在不同的缺失率下通过回判,得到了半参数方法与BP神经网络的准确率,并将两种方法分别运用到不完全的脂肪肝临床数据中。结果表明,两种方法都适用于处理小样本情况下的缺失数据问题,当缺失率较小时基于BP神经网络的准确率较高,当缺失率不断上升时,半参数方法的处理结果比较稳定。The absence of data not only increases the complexity and difficulty of statistical analysis,but also leads to bias in analysis results.In this paper,the advantages and disadvantages of the semi-parametric method and the BP neural network method are compared in the case where the component index data conforms to the normal distribution and is a random missing mechanism.The iris dataset for simulation study and pass the judgment at different missing rates are selected.Then the accuracy of the semiparametric method and BP neural network is obtained.Finally,the two methods are applied to incomplete fatty liver clinical data separately.The results show that both methods are suitable for the problem of missing data in the case of small samples.When the missing rate is small,the accuracy based on BP neural network is higher.When the missing rate is rising,the processing result of semi-parametric method is more stable.

关 键 词：半参数方法 BP神经网络 随机缺失机制 

分 类 号：O29[理学—应用数学]