动态粗集理论在决策树算法中的应用研究

RESEARCH ON APPLICATION OF DYNAMIC ROUGH SET THEORY IN DECISION TREE ALGORITHM

基于动态粗集理论,提出一种改进的动态粗集决策树算法。改进后的算法对每一属性计算数据对象迁移系数的和,值最大的属性成为决策树的根;在对决策树分叉时,给每一决策类别的数据对象集合计算从根到分叉属性所构成的属性集的膨胀度,值大的属性构成分支结点。算法在UCI机器学习数据库原始数据集及其噪音数据集上的实验结果表明,该算法构造的决策树在规模与分类准确率上均优于ID3算法及C4.5算法。

Based on the theory of dynamic rough set, we present an improved dynamic rough set decision tree algorithm. The improved algorithm calculates for each attribute the sum of transition coefficient of data object and the attribute with maximum sum value will be chosen as the root of the decision tree. When bifurcating the decision tree, the algorithm calculates for the collection of data objects in each decision category the expansion degree of the attributes set consisting the attributes from root to branches, and the attributes with maximum expansion degree will be select as the branch nodes. The experimental results of the algorithm on primary dataset of UCI machine learning database and on generated data sets with noise points proves that the decision tree constructed by this algorithm achieves better classification accuracy and smaller scale than those of ID3 and CA. 5 algorithms.