Selection Methods Of Attributes And Optimization Of Method In Rough Sets

Rough sets theory proposed originally by Z. Pawlak is a new data mining method, which has been successfully applied in many fields, such as feature selection, pattern recognition, machine learning, decision analysis, etc. Data mining based on rough sets technique is accomplished by using attributes reduction algorithm, which can be roughly classified into three categories:forward reduction algorithms, backward reduction algorithms, and other reduction algorithms (for example the attribute reduction algorithms based on discernibility matrix).The reducts obtained by forward reduction algorithms usually retain some dispensable attributes for classification; this paper has improved the HORAFA algorithm to remove the superfluous attributes. Firstly, an attribute with the lowest significance is deleted from the discernibility matrix. Secondly, matrix unit with one attribute only is added to the reduct step by step, finally a reduct without dispensable attributes can be obtained. For backward reduction algorithms, some equivalence classes can be obtained by removing an attribute, the properties of the equivalence class;are analyzed, a novel backward reduction algorithms with halt criterion based on cardinality of partition is proposed, the numbers of decision rule converted from the obtained reduct can be decreased. A new concept of set cover is presented, which is the generalization of traditional set cover, and we integrate the new concept in attribute reduction algorithms based on disernibility matrix.When we use the forward reduction algorithms to reduce the dispensable attributes, the previous results must be repeatedly computed, in order to overcome the drawback an iteratively methods is proposed. We firstly partition the instances by using the decision before we construct the discernbility matrix, this can effectively decrease the time for constructing the discernbility matrix.