Abstract: An algorithm of the decision tree based on class feature matrix is proposed.This algorithm regards the core attributes as a beginner， and fully takes the property of the discernible matrix and the importance of each attribute into consideration.It also uses the class feature matrix to get the reduction decision table and mine the decision rules， and finally creates the decision tree according to the scattering rules.The comparable and analyzable experiment shows that this algorithm can make a minimize decision tree whose rules are true.

Key words: rough set, class feature matrix, decision tree, decision table

摘要： 提出了一种基于类别特征矩阵的决策树算法。该算法以决策表的核属性为起点，充分考虑了可辨识矩阵的特性和单个属性的重要性，利用类别特征矩阵对决策表实现最简化决策表的确定和决策规则的挖掘，最后实现最简规则的决策树生成。通过应用实例比较分析，证明该算法能生成最小化决策树，并且决策树生成规则切合实际。

关键词: 粗糙集, 类别特征矩阵, 决策树, 决策表