Research On Decision Rules Extraction Based On Rough Set Theory And Decision Tree

Rough Set theory is a kind of effective analysis tool to deal with inaccurate, incomplete, and uncertain information, which makes use of the existed rules in the knowledge warehouse to character the incomplete information. Attribute reduction and decision rules extraction are two main respects research area of rough set, but there are many defects in the existing the two algorithms.In order to obtain a more streamlined set of attribute reduction effectively, the paper optimizes the attribute reduction algorithm that based on discernibility matrix firstly. Because there are some defects, such as element duplication, complex calculation, and varying length of discernibility matrix element in constructing the traditional discemibility function. As the decision tree technology is characteristic with fast classification speed, high efficient, easily to be understood, and so on, the paper combines the decision tree and rough set theory to extract decision rules, in which, the optimized attribute reduction algorithm is applied to get reduction sets and then the reduced attribute set is used to construct a multi-variable decision tree to extract decision rules. At last, in order to avoid disturbance of inconsistencies information, the accuracy and coverage degree are introduced to filter the decision rules and extract decision rules effectively. An example of rotating machinery fault diagnosis validated the above optimized algorithm, which shows the methods combines rough set and decision tree can not only wipe off noise, but also deal with inconsistencies information.In order to put the above optimized method into practice, a decision rule extraction system based on rough set theory and decision tree is developed in this paper. The system is designed based on .NET platform, which can carry out attribute reduction for original decision table, extract decision rules according to the structured decision tree, and obtain the effective decision rules finally.