摘要
应用粗糙集理论,提出了一种新的多变量决策树构造算法.该算法以核相对于决策类的泛化来划分样本集,如果所划分子集的样本存在不一致决策类并且未用于划分的属性为空时,试探着分别把该子集和一致性子集合并,计算各合并子集的条件类对决策类的确定性程度,选择确定性程度大的作为同一子集,并用一致性子集的类标号进行标示.和苗夺谦提出的多变量决策树算法比较,本算法充分考虑了训练集中的噪声数据,允许在构造决策树的过程中划入正域的实例类别存在一定的不一致性,可简化生成的决策树,提高决策树的泛化能力.
In this paper, a new method to build multivariate decision trees based on Rough Set is proposed. The relative generalization of core with respect to decision attributes is used for devising to the decision table, if a certain inconsistency exists in the subset of examples and there are not attributes, then the inconsistency subset merged with the consistency subsets, respectively. And the measures of certainty of condition equivalence relation with respect to decision equivalence relation are accounted respectively, the great result of the combined subset is got, and the label of the consistency subset to label the node. Compared with the algorithm of Literature E51, noisy data of training sets are considered. A certain inconsistency is allowed to exist in examples of the positive regions, so the method can simplify the decision trees and improve its extensive ability.
出处
《洛阳师范学院学报》
2011年第5期74-77,共4页
Journal of Luoyang Normal University
基金
河南省教育厅自然科学研究计划项目(200913520021)
河南省科技发展计划项目(102102210472)
关键词
多变量决策树
粗糙集
相对泛化
确定性程度
multivariate decision tree
rough set
generalization
measure of certainty