A new approach of the division optimization of attribute spaces for the decision tree construction

The reasonable division of attribute spaces is a core problem in the decision tree construction and the rule extraction, which directly influences the effectiveness of the construction of decision trees. In this paper, a new two-dimension analysis method for the attribute space division is proposed, which not only reduces the analysis complexity, but also improves the efficiency of attribute division. And a new model for optimizing the division of attribute spaces is proposed. The new model can effectively measure the reasonability of the boundary of attribute spaces and satisfy the condition with maximum and uniformity of the boundary distance of the separation of attribute spaces. Furthermore, this paper gives the concept of non-discriminate regions, and presents the method for distinguishing the non-discriminate regions and the region collision and the approach of the separation of attribute spaces. The typical computing examples prove the validity of our approach in attribute division and decision tree construction.