Incomplete Information System Based On Rough Set Attribute Reduction Method Of Research

In the real life, because of the unpredictability of the external environment and thelimitation of human thought, this may lead to data loss and the poor quality of data,especially the loss of reliable data will result in lack fidelity of data. The process of rulegeneration might be in chaos and generating unreliable data due to the data loss inincomplete knowledge system. In the recent years, classical rough set theory proposedby Pawlak shows the biggest potential in the aspect of knowledge acquisition incomplete information system. Nowadays there are two methods to deal with incompleteinformation system. One method is data completion, namely on the premise of notchanging the quantity of information of information system the loss data is completed.The other is rough set extension model of incomplete information system. Classic roughset classifies objects by indiscernibility relation. However equivalent relation is thoughtto be much more rigorous in incomplete information system, data processing is done byextending the limitation of equivalent relation.The main content of this dissertation is about the research of attribute reductionbased on the rough set incomplete information system, including the extension of roughset theory and attribute reduction of incomplete data table. And it can be generalized asfollows:(1) Rough set theory and the current situation of incomplete information system inrough set theory are respectively presented. And important tasks to be carried out the inthe intersection of rough set theory process are shown. Rule generation of mass data,processing method of loss data, the efficient way of attribute reduction and theintegration of multi-field are included. The result of data loss in incomplete informationsystem, the definition of null value, the common ways of data completion and theimportance of incomplete information process are introduced.(2) The rough set extension model of incomplete information system undertolerance relation, similarity relation, valued tolerance relation and limited tolerancerelation is proposed. And corresponding examples are given to discuss the advantagesand disadvantages of each model. On that basis, a new rough set extension model is put forward. In this new model, threshold value is imported in the concept of degree ofapproximation. So users can reasonably set the value of according to the dispositionof null value in data table. And this fits the practical application of knowledgediscovery.(3) The differences between the process of attribute reduction in completeinformation system and the process of attribute reduction in incomplete informationsystem are compared and analysed. Data loss may bring about incompatibility of thedecision data table. And the discernibility matrix in complete information system alsocan’t deal with this problem very well. As a result, combining knowledge granulation,consistency description is renewedly given. This overcomes the incompatibility andreduces the comparison of missing value in discernibility matrix. At the same time,combining new model and heuristic knowledge, Based on the quantity of information inthe attribute reduction algorithm。(4) The algorithm of decision information reduction is come up with.Inconsistencyin the model of similarity limited tolerance relation is used to describe the quantity ofdecision information. The less the inconsistency is, the more the quantity of decisioninformation is. Bonding with new model, the quantity of information in decision tablebetween the objects is depicted by adjusting the value of. The significance ofattribute is redefined as the heuristic knowledge to do attribute reduction.