处理高杠杆异常值的抗隐差型Bayes方法

Multiple Outlier Detection in Observations Including Leverage Points with Outliers by Bayes Method

摘要: 给出了一种剔除初始子集中高杠杆异常值的方法。首先根据高杠杆异常值在总观测值集中所占的比例选出若干组观测值,使得至少有一组不含高杠杆异常值的概率很高;然后根据残差最小准则从中选出不含高杠杆异常值的那组作为初始子集;最后用这种初始子集确定方法结合Gibbs抽样给出了相应的Bayes多粗差定位算法。

Abstract: A new method concerning how to eliminate leverage points with outliers from the conditional subset is considered.Firstly,several subsets of observations are selected randomly from all observations in accordance with the percentage of leverage points with outliers in order to obtain one subset containing no leverage points with outliers.Secondly,the conditional subset is selected from the subsets that are selected above according to the sum of squares of the residuals.Lastly,a new algorithm is given by connecting our procedure with the Gibbs sampling.