Research And Application Of Data Reduction Algorithms Based On Rough Entropy

Rough set theory is a math theory which processes the non-accurate after probability theory, fuzzy theory and evidence theory. The reduction of knowledge is the main topic in rough set theory. Getting the best reduction or all reduction is the NP problem. Heuristic algorithm based on the attribute importance had been made to get better reduction.This paper defines the importance of essence through roughness of knowledge followed by the description of roughness of knowledge and the suggestion of definition of rough entropy and establishment of the dependency relation between knowledge and rough entropy in the information system. Accordingly, the contrast of dependency is defined, and a new significance of attribute is defined based on the new rough entropy of knowledge. The judgment theorem with respect to attribute reduction is obtained, and it is the same with not only general decision information system but also information system, and provides theory basis for knowledge roughness in information systems. The paper offers algorithm for reduction of knowledge based on rough entropy of contrast of dependency. It proves that the algorithm is effective through example analysis.To represent complete algorithm for attribute reduction in rough set theory, based on rough entropy of knowledge, A new conditional rough entropy of decision concept sets is proposed. The conclusion that conditional rough entropy of knowledge decreases monotonously as the information granularities become finer was obtained, and a heuristic algorithm was proposed. Theoretical analyses and experimental results prove the validity of this reduction method.Value reduction is rule reduction in essence. First, the relation is established between knowledge and rough entropy in the information system. Based on the degree of dependency of the attributes,which is often taken to describe two distributions, a new rough entropy of knowledge is defined and an new algorithm for acquisition of decision rules by hierarchical reduction is proposed, and the reliability and coverage degree for rules are also introduced to analyze the approximation degree. The experiment and comparison show that the time complexity of this algorithm is less than that of matrix computation for rule extraction which is in inconsistent decision tables, and the algorithm provides more precise and simple decision rules efficiently.Finally, a verification system is configured to validate our proposed algorithms, and achieves the reduction of meteorological information system.