Research On Heuristic Attribute Reduction Algorithm Based On Rough Set

As the network development and popularization of the rapid growth of data, in order to obtain useful information, traditional data mining technology is constantly changing, which deal with the uncertainty of rough set based approach will be applied to the rough set data mining, due to its unique advantage that no outside information in any collection of data processing, which ensures that the problem description and processing of objectivity to a certain extent, thus becoming a mainstream method. Attribute reduction as its core, the essence is to remove some of Knowledge irrelevant or unimportant redundant attributes, so that through the knowledge base after the reduction in the amount of data processing dimension can be greatly reduced, or for subsequent classification rules extraction facilitate the work, while maintaining the ability to classify the original knowledge base or decision-making capacity unchanged. People often want to be able to delete those as much as irrelevant or unimportant attributes from the information table to obtain the minimum reduction. However, it is not the presence of a very effective way to get the minimum reduction, therefore, hot and difficult rough set theory is looking for a quick and effective reduction algorithm and the minimum reduction, which has great theoretical and practical significance.Through in-depth analysis of rough set theory and classical attribute reduction algorithm based on this work presents the main two aspects:First, to get a better combination of attribute reduction in decision table, from the information theory point of view, on the basis of distinction based matrix, to an improved conditional entropy reduction algorithm for heuristic information. The algorithm conditional attribute with respect to the decision attribute conditional entropy and the distribution of property values to be considered simultaneously. Then use them as inspiration factor than regives a measure of the importance of property basis, and eventually set attribute reduction. The experimental results show that the algorithm can attribute reductions, and the reduction results obtained most concise decision rules combination.Secondly, incomplete decision table, this paper combines the concept of conflict domain, first quickly find core attributes based on the number of objects in the domain of conflict changes, and then generate the core attributes targeted for reduction of different ideas, if the presence of the core attributes reverse verification to determine if the core attributes is the final result, on the contrary, the number of objects by attributes importance as a collision domain reduction algorithm determines the conditions, determined the final reduction. According to the same time in both cases we consider designing an efficient reduction, and show that this algorithm is feasible.