Research On Variable Precision Rough Set Model In Data Mining

Since the Rough Set has been presented, its theory developed continually. In some aspects Rough Set and method get overcomes many deficiencies in traditional data analysis, so it has been given extensive attention at home and abroad. But a limitation of classic Rough Sets Model is that the classification which it deals with must be totally correct or definite, Because it classified according to the equivalences strictly, so its classification is accurate, namely"include"or"not include",but the one that does not have to a certain extent"includes"or"belongs to". In order to deal with this kind of limitation Professor W.ziarko developed the Variable Precision Rough Set Model (VPRS) in 1993.The main research of this paper is as follows:(1)It analyses the reduction characteristic of RS and VPRS: quality of classification, relative positive region and lower approximation. Compare the difference of RS and VPRS.(2)By introducing the inclusion proportion threshold value for each condition class, it makes a lucubration on reduct anomaly about variable precision rough set model, and describes the range relation between inclusion proportion and quality of classification. Then it analyzes the reduct anomaly when inclusion degree vibrates and positive area changes. The basic ideas are presented to eliminate reduct anomaly. At the end it gives the range reduct definition, and realizes the range reduct algorithm. All of this develops the reduct of the variable precision rough set model.(3) The research of the 4th chapter mainly focuses on knowledge reduction based on VPRS model.firstly,it draws up two algorithms: the approximate reduction algorithm and distribution reduction algorithm,analyzes them and points out their strongpoint and shortcoming. Secondly, it introduces an improved algorithm which revises the discernibility relation of similarity sets on the basis of the Discernibility Matrix of distribution reduction. The improved algorithm eliminates the harsh requirements of distribution reduction. To some degree, it overcomes the drawback of possible reduction that the derived decision rules may be in full incompatible with the ones derived from the original system. Finally, theoretical analysis and experimental results are used to prove the improvement of the algorithm.