Abstract: Aming at irrelevant and redundant attributes could decrease the classification accuracy of decision tree, this paper proposes a method that building decision tree based on the reduction that chosen base on the consistency criterion. After the process of discretization for continuous attributes for the 5 twoclass samples from UCI machine learning repository, constructs C45 and CART decision trees based on rough set theory and consistency criterion respectively. The experiment based these 5 samples shows the method based on consistency criterion is efficient and feasible for decision tree building.

Key words: rough set, attributes reduction, decision tree, consistency criterion