一种面向纵向数据的RE-BET算法及应用  

A RE-BET Algorithm for Longitudinal Data and Its Application

作　　者：鲁庆[1] 穆志纯[1] 

机构地区：[1]北京科技大学自动化学院,北京100083

出　　处：《科学技术与工程》2015年第16期77-83,共7页Science Technology and Engineering

基　　金：国家科技基础性工作专项基金(2012FY113000)资助

摘　　要：混合效应模型是分析纵向数据的有效方法,但模型的线性结构限制了其适应现实数据的能力。提出了一种RE-BET算法及其变形的RE-BEBT算法,采用树形方法估计混合效应模型的固定效应,可以自动选择重要变量,能更好地发现和描述变量间关系;采用基于Dirichlet过程先验的贝叶斯方法估计混合效应模型的随机效应,使模型可以适用于小样本数据。以低合金钢和碳钢的海水腐蚀数据为例,通过与实验数据和其他算法的计算结果对比分析,验证了RE-BET算法可行性和有效性。Mixed effects model is a useful methodology for longitudinal data, but the linear structure restricts the model ability to handle the data from real world, a RE-BET algorithm and its deformation named RE-BEBT were proposed. The algorithm used tree-based method to estimate the fixed effects of mixed effects model so that it could select important variables automatically and could discover the relationship between variables. In order to apply model to small sample size data, a Bayesian method based on Dirichlet process prior was used to estimate the ran- dom effects of mixed effects model. The RE-BET algorithm was appeied to the corrosion data of low alloy steel and carbon steel in Seawater and compared the result with other algorithms or experimental data, showing that the RE- BET algorithm is feasible and effective.

关 键 词：纵向数据 Dirichlet过程 树形方法 腐蚀数据 贝叶斯 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]