Based on the rough set theory,this paper considers the contribution of each partitional attributes to entire classification,and gathers the contribution to avoid local best effect.In this foundation,The authors can integrate variable precision model and replace approximate precision with variable approximate precision.Then offer a new concept of variable precision branch summarized roughness which is regarded as the criteria for choosing attributes when constructing a decision tree.It improves the accuracy of attribute selection,overcomes the impact of noise data effectively and strengthened g...