基于异常检测的NVP表决器

一般的表决器多是针对一维输出进行表决,构建表决器相对应的决策树,预先给出阈值来制定表决策略。但是对于多维输出的表决器研究相对较少,尤其是输出各个变量之间非独立的情况下,不能将多维输出的每个变量进行单独表决,否则会引入异常点,影响表决结果。实验表明,将异常检测技术引入到表决器中,净化需要表决的数据,可以提高表决结果的准确率。

NVP Voter Based on Deviation Detection

At Present the majority of voters are for one-dimensional output,which build decision trees and voting strategies according the parameters gave.Meanwhile,study for multi-dimensional output is relatively few.Especially in the case that the variables are related each other,we can’t vote for each of the variables separately,otherwise the deviation will affect the voting result.Experimental results show that deviation detection technology can clean the data set which will be voted,and improve the accuracy.