基于叶枝比率的决策树构建算法  

Algorithm of Constructing Decision Tree based on the Leaf and Branch Ratio

作　　者：刘军[1] 

机构地区：[1]南京工业大学电子与信息工程学院,江苏南京210009

出　　处：《信息网络安全》2013年第2期9-12,共4页Netinfo Security

基　　金：国家自然科学基金[60673185];教育部留学回国人员科研启动基金[200711108]

摘　　要：当前构建决策树算法中主要存在准确度不高与计算量大两个问题。文章认为影响准确度的因素是属性作为基本判定条件权重过大;影响计算量的因素是采用了概率计算方法。针对以上问题,文章提出了利用叶枝比率判定算法构建决策树:即将属性按取值划分为若干叶(基本粒),而将条件属性叶所对应的决策属性叶的类别数量定义为枝,以粒分辨关系为理论基础,用全局叶枝比率判优法直接确定划分属性,由此自顶向下构造决策树。理论与实例分析的结果表明,利用此算法所建决策树准确性高且判定的算法简洁。There are two problems in the construction of the decision tree algorithm which are accuracy and computation complexity. It＇s considered that the influence to accuracy is due to the attribute as a basic determined factor and the impact of the computational complexity is due to use the probability theory. In accordance with above two problems, the paper put forward the algorithm of construction decision tree based on the leaf and branch ratio： the attribute according to the value is divided into several leaves （basic granulatio ）, the condition attribute leaf corresponding to the class number of decision attribute leaf is defined as the branch. The algorithm is based on the definite relation of granulation. The partitioning attribute is directly determined by the global leaf and barnch ratio and a decision tree can be built form top to down. Theory and example analysis results show that the algorithm used in building decision tree is accurate and simple.

关 键 词：粗糙集 决策树 粒分辨关系 叶枝比率 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]