基于粒划分方法构建决策树的算法研究  

Research on Algorithm of Constructing Decision Tree Based on Granulatio Division Method

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：《计算机技术与发展》2012年第10期87-90,共4页Computer Technology and Development

基　　金：国家自然科学基金资助项目(60673185);教育部留学回国人员科研启动基金资助项目(教外司留[200711108号])

摘　　要：针对当前基于信息增益和粗集属性约简作为属性选择标准建树算法存在的不足,以粒划分方法为理论基础,将属性按其取值划分为若干属性粒,提出以属性粒的长度量和其所对应决策属性的粒类别个数作为确定分裂属性的基本参数,自顶向下逐级构造决策树,不涉及信息增益、等价类和属性约简等复杂运算的中间过程。该算法的优点在于不仅考虑本层结点的划分而且预测下层结点的走向,具有较高的精准度,而且解决了当前建树算法不具有普遍适应的难题。In view of the current algorithm building decision tree has shortcomings by attribute as selection standard based on information gain and attribute reduction of rough set, the attribute according to the attribute value is divided into a number of granulation by the granu- lation division method as theoretical basis, put forward attribute granulation length and the corresponding decision attribute granulation class number as basic parameters of determined splitting attribute and stepwise construct decision tree from top to down. The algorithm does not involve complex operation process with the information gain, equivalence classes and attributes reduction. The advantage of this algorithm is that not only considers the layer nodes division but also predicts the trend of lower nodes ,has high precision and solves the problem that the current algorithm of building tree is not universal adaptation.

关 键 词：粗糙集 决策树 粒划分 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]