The Research On Data Mining Algorithms Based On Rough Set And Computational Intelligence

Massive data and lacking of knowledge led to the appearance of data mining technology which main goal is to obtain effective, implied, previous unknown and valuable information. Today data mining is in the forefront of research topics of database fields. The rough set theory is a useful tool to deal with vagueness and uncertainty. It can induce decision or classification rule through knowledge reduction while the classify ability was not decreased. The technology of computational intelligence which includes neural networks, fuzzy logic and evolutionary computation imitate human way of thinking and evolution and has been widely used in the industrial control, pattern recognition, and other fields. Currently, in the area of data mining the combination of rough set and computational intelligence is mainly used in the stage of data preprocessing and rarely used in classification, clustering and the mining of association rule algorithm.The basic concept and application of data mining, the theory of rough set and the theory of computational intelligence are introduced. The application and mutual complement in data mining of rough set theory and computational intelligence are analyzed. In view of the predominance of rough set theory in dealing with symbolic attributes, a new clustering algorithm is proposed which use niche genetic algorithm based on sharing mechanism to divisive hierarchical clustering and use the theory of rough set to define fitness function of GA. The experiments show that the clustering accurate rate of this algorithm is higher than another two algorithms for symbolic attributes. In addition, a new classification algorithm is proposed which combines rough set theory and organizational coevolutionary algorithm for classification and use the idea of support subset to determine the fitness of organizations and use the suggestion of the support subset to accelerate the evolution and enhance the synergies between the groups. Experiments show that when all attributes of data set are symbolic attributes this algorithm has higher classification accurate rate and costs less time than another algorithm.