Research Of The Rough Set Model Based On Relation And Its Data Mining Methods

The rough set whose application principle is to distinguishing knowledge based on the distinguished relation and getting the knowledge reduction was proposed by Pawlak.Then we can get the internal laws and values of knowledge. Rough set theory has outstanding effects in dealing the uncertain and incomplete problems. It also has successful application in data mining, machine learning, fault diagnosis, and so on. Although the theory study of rough set is largely improved, for the new problems emerging, we still need to further study in extend rough set model.In this paper, firstly, the rough set model of variable precision probability dominance relation is proposed. The new model has a certain degree of fault-tolerant ability. It can reduce the sensitivity of data noise and the complexity of problems through the study for the definitions, theorems and examples of the model. Secondly, the improved rough entropy is proposed which is more suited to address the uncertainty measure based on roughness and rough entropy. The new entropy not only solves the problem of uncertainty in the model based on variable precision probability dominance relation, but also covers the shortage that existed in the method of roughness and rough entropy. Finally, three attribute reduction methods are analyzed to do the data mining. The first one is analyzed by using the existed dominance degree in the model based on variable precision. Then we can use it to ensure the attribute importance and get attribute reduction. The second one is the improved conditional entropy. It can solve the not monotonic problem of conditional entropy in incomplete system,so it can increase the efficiency of the algorithm reduction. The last, a reduction algorithm based on new knowledge granularity is proposed by combining new model. It proposes a new method to do attribute reduction. The method proposed in this paper provides new theoretical approach for the data mining.