Research Of Some Intelligent Mining Algorithms Based On Knowledge Roughness And Extended Reducts Of Rough Sets

Rough sets theory is an effective approach to analyzing incomplete and uncertain information systems. Rough sets analysis only uses the internal information of data and does not rely on any prior knowledge . It has been successfully applied to data mining and KDD, which promotes itself as a predominant approach in DM and KDD.The classical rough sets model was mainly set up for dealing with complete information systems without any missing data nor with imprecise data and such systems have been the topics for many rough set based knowledge reduction. However, we often need to deal with incomplete or inconsistent data in practical applications because of some undesirable reasons , which puts forward too many problems to the classical rough sets to solve . So, many extended rough sets models merge as the time requires ,such as the variable precision rough set , fuzzy rough sets etc . The extended models of rough sets introduce approximate and probabilistic ideas into classical rough sets and define new reducts , core and discernibility matrix of themselves and have made some satisfactory achievements in certain applications.In this paper, we first propose a new approach to the construction of a decision tree based on knowledge roughness and the decision tree obtained by this algorithm is simpler than that of ID3.In the third chapter, we analyze the properties of some reducts in inconsistent information system , namely the distribution reduct, the maximal distribution reduct and the reduct under entropy perspective. As a result of an intensive analysis of them , we conclude an equivalence between the distribution reduct and the one under entropy theory.Furthermore , we bring forth a new way to measure the casualness of the relationship between the reduct attributes and the decision ones withthe help of chi-square in statistic .which provides referenced information when choosing a best reduct from among many candidate ones. In many special applications , dynamic additions and modifications of data require learning algorithms to learn incrementally on the base of previous knowledge instead of running the mining algorithm from scratch every time. Therefore , we present an incremental learning algorithm based on the maximal discernable matrix and we apply the algorithm to the clinical data analysis .The learning results prove the effectiveness and correctness of the incremental learning algorithm in the end.