Data Reduction And Rule Extraction Based On Rough Set

Rough Set theory is a mathematics tool for processing vague and imprecision knowledge, which is putted forward by 1982 from Poland scientist Z.Pawlak. Rough Set theory has already been successfully applied in the area of Machine Learning, Pattern Recognition and Data Mining etc. The most important point on using Rough Set theory to mine data and extract rules from knowledge is study the algorithms of attribute reduction and rule extraction based on Rough Set theory. To reduce the attribute's dimension and extract the knowledge rules adapted to decision support by reduction is one of the important applications of Rough Set theory.Attribute reduction is NP-Hard problem, the main reason, which causes it, is attribute combination explodes. At present, however there does not exists an effective method to do with it, exploring the speedy attribute reduction algorithm therefore still is one of Rough Set theory research hot spot. Firstly this paper has studied CEBARKCC, and improved the CEBARKCC by diminishing unnecessary calculation of the importance degree of attribute. The efficiency of the algorithm has been improved. Secondly, a new algorithm is putted forward, which is based on some property of probability-distribution function in Rough Set theory. The algorithm turns to start at full attribute set and circulation deletes redundancy until there isn't redundant attribute can be delete, and at last gets attribute reduction. As the algorithm has avoided the complicated process for calculating the important degree of attribute or the core of attribute, less calculation time is need. The efficiency of the calculation is even higher than that of discernibility matrix and entropy of information.On rule Extraction, a rules extraction algorithm of decision tree based on Rough Set theory is putted forward in this paper. The node of decision tree which the algorithm choices is the attribute which not only causes the confidence degree of decision rule to satisfy the confidence degree which the user input but also makes the rule have biggest support degree which is based on the definition of supporting degree in probability-distribution function . This method simplify the production of decision tree, noise rules are eliminated. Decisional rules which are extracted are more precise and simple. Finally, this paper implements the three algorithms. The validity and feasibility of the algorithms are demonstrated by large amount of data in the UCI machine learning databaseThis research is supported by science and technology item"Knowledge Acquiring Based on Grey Rough Set Theory"(Ja05290)of Fujian Education Community.