Attribute Reduction Algorithm Oriented Rough Set

As a new soft computing method, Rough Set Theory deals with the uncertainty and imprecision problem. It provides a very effective theoretical framework for machine learning, data mining, knowledge acquisition, decision analysis, pattern recognition and expert systems in areas such as information processing. This paper systematic elaborated the basic theory of rough set and under the support of the basic theory framework, mainly done the following several aspects of research:Firstly detailed analysis and research the classic attribute reduction algorithm, focusing on the summary and comparison of the attribute reduction algorithm between the decision tables that there is nuclear and there is no nuclear, it found that the traditional algorithms are not suitable for there is no nuclear attributes of decision table. Based on the above problems, this paper proposes a attribute reduction algorithm that based on the clustering rate, this method is based on the attribute degree of differentiation, with the same attribute degree of differentiation, then use clustering rate fixed attribute importance, to ensure the inevitability of the existence of starting point attributes, thus completing the starting attributes calculating and the attribute reduction of decision table.However attribute reduction algorithm based on clustering rate is always on the premise of discernibility matrix, it will consume large amounts of complex data storage space for store and at the expense of the computing complexity when involving a large number of calculated data. On this basis, it draws lessons from the idea of tree diagram, and put forward a kind of tree graph based recursive attribute reduction algorithm in order to bypass the disadvantages of discernibility matrix. And it is theoretically proved that the algorithm is not only suitable for nuclear decision table, but also suitable for soft decision table. Experiments prove that the algorithm shows advantages in dealing with large data sets that have no nuclear attributes. Then put improved reduction algorithm into the formal concept analysis will not only saves the storage space greatly, more important is a simple tree graph replaces multifarious concept lattice, also attribute reduction process and the concept generation process almost simultaneous.Finally, design verification experiments are analyzed. Test and evaluate from the simple level of attribute reduction and classification accuracy of reduction subset.