Research On Attribute Reduct And Rules Acquisition Of Inconsistent Decision Tables

In the real world, because of the definition and standard of the data collection and operation and other problems, and the decision tables are often inconsistent. How to eliminate or minimize the negative impact of inconsistent decision tables has become an important research topic in knowledge discovery and data mining of information systems. There two key issues:attribute reduction in inconsistent decision table and the rules acquisition in inconsistent decision tables.In this paper, inconsistent decision table is considered as researching object, rough set theory and genetic algorithm are considered as researching method. This paper does some researching on attribute reduction and rules acquisition of inconsistent decision tables respectively. First, the paper describes the four kinds of attribute reduction of inconsistent decision tables:algebraic reduction, distribution reduction, assignment reduction and maximum distribution reduction, and discusses the relationship among the four. Because the distribution reduction misses the information for the least, so this paper focuses on it. Then, based on existing research of the classical discernibility matrix, we propose a new discernibility matrix method which can get distribution reduction. This method not only can accurately obtain the distribution reduction of inconsistent decision tables, but also greatly reduce the storage space of a discernibility matrix. Finally, we present a new rules acquisition method based on rough set theory and genetic algorithm RSGA. Using rough set theory, we will divide inconsistent data table into two parts, certain data and possible data, and then standard genetic algorithm is used for mining rules set. When the algorithm is processing, the user is allowed to set three evaluation parameter values of the rules:support, confidence, coverage for specific application needs. This algorithm will delete the rules which do not meet the requirements, so we can reduce the amount of data in the case of massive data. The advantage of this setting is obvious. Using the proposed two new methods, we will handle inconsistent decision table problem effectively:minimizing the loss of information in the case of simplified decision table, then digs out certain rules and possible rules by using RSGA algorithm.