Research To Data Mining Of Incomplete Information System Based On Rough Set And Neural Network

In practical issues, data in database is seldom complete. Data waiting to be processed is incomplete to some degree. Incomplete Information System exists at large, so it is unreasonable if we directly apply data mining methods for complete information system to incomplete information system. And it will badly influence the effect of mining. So, it is important that researching how to do data mining on this kind of incomplete information system.Incomplete information is caused by many reasons. Under practical circumstances, some data can't be observed because of practical limits or data lost caused by maninduced factors or storage medium's failure, transmission medium's failure or data hidden. There is mainly two explain, data missing and data absent. Value missing can be gained again in the future, but data absent can not be gained again.Rough Set is a method of apply to research incomplete, indefinitely knowledge and apply to expression, studying, concluding data. The main thinking of the rough set is educing the decision of questions. Neural Network has stronger learning ability. It could learn automatically from training instances based on certain learning algorithm. Each of Rough Set theory and Neural Network has predominance and disfigurement when process the information. We combine those two algorithms to make their predominance is reinforced each other. First, we transform incomplete information system into complete information system by Rough Set theory, and then process data mining for the completed system by Neural Network method.We bring forward an improved ROUSTIDA algorithm. We consider filling of missing values should reflect the basic characters and the connotative internal rules of the information system. Based on the Distinguish Matrix, we expend the using range of original algorithm, transform incomplete information system into complete information system, and make it more reasonable and effective.We bring forward that make the impact degree to numerable of which the condition attributes of information system object impact to decision. According to the impact degree for the consistency, find the best consistency object to the object which has missing values, and set up a consistency token filling method.Finally, use the Neural Network method to learning and simulation the completed information system. It has been proved by experimentation that the completed method is reasonable and effective.