数据污染对于序贯分支方法的影响及应对策略

The influence of data contamination on sequential bifurcation and its counter measures

序贯分支方法(sequential bifurcation,SB)因其高效性,近年来被广泛用于仿真试验的因子筛选研究中.然而,传统的序贯分支方法难以应对数据污染情形下的因子筛选问题,因此,本文结合稳健估计的方法改进了传统的序贯分支筛选过程,使其具有良好的抗异常值特性,解决了多种数据污染情形下的因子筛选问题.首先,分析仿真试验中可能出现的数据污染情形及其数据形式,并结合序贯分支方法的基本原理,量化不同数据污染情形对因子筛选结果所造成的影响;其次,采用稳健的位置和散度统计量改进了传统的序贯分支方法中的显著性检验过程,使因子筛选结果不受数据污染的影响;最后,通过仿真试验验证改进的序贯分支方法具有更好的抗异常值特性,同时,该方法在非数据污染下也不失一般性.

Sequential bifurcation(SB) is the most efficient factor screening method which is widely used in the simulation experiments to identify important factors with as few runs as possible.However faced with data contamination problem,SB which assumes the normal response shows some limitations.Hence,classic sequential bifurcation procedure is modified in this paper,which could handle not only the response with data contamination but also the normal ones.Firstly,several types of data contamination are introduced and the influence of factor screening results caused by data contamination is analyzed.Then sequential bifurcation is modified by including robust estimators in hypothesis testing.Finally,Monte Carlo simulation is employed to show that the modified sequential bifurcation method is efficient and effective when the data are normally distributed yet also robust when the data are contaminated.