Classification Research Based On Evidence Theory And Covering Rough Sets

Classification is the most important issues of data mining. Many data mining issues can be taken as classification issues equivalently.Both of evidence theory and rough set theory are important methods of dealing with uncertain issues. Evidence theory uses basic probability assignment fuction,belief function and plausibility function to deal with uncertain issues,demostrates its advantages on classification data mining area.The advantage of rough set theory in dealing with uncertain issues is that it does not need any previous imformation,and it's widely used on classification.However,In many pratical classification issues, the relation between objects are covering relation,which hampers the application of Pawlak rough set model and could not indentify the category of samples effectively.To keep some of the good properties of Pawlak rough set model, there is necessary to transform the covering rough set into Pawlak rough set.For rough set theory can not contain mechanisms of dealing with imprecise or uncertain data,it is a strong complement of evidence theory,and can solves many problems that evidence theory brought in.The article first introduces the background and significance of evidence theory and covering rough set classification,the research status of evidence theory, the research status of Pawlak rough set, the research status of covering rough set,the research status of the relation between evidence theory and covering rough set.Then it introduces the relative knowledges of Pawlak rough set,covering rough set and evidence theory.The article studies on neighborhood covering rough set and the equivalence relation which is derived by covering rough set,proposes a new transformation of covering rough set,the new transformation proves that a covering can only derive a partion,a covering rough set and its reduct derive the same partion. It proves that covering approximation operators of the derived equvialence relations are less coarser,the approximation degree of a set could be improved.The article proves the superiority of the transfermation by instances.The article researches on all kinds of pignistic probability transfermations and decision rules of evidence theory.It exploits a programming method to gain compatible probability of the category of samples and compares with other pignistic probability transforms, then it makes decision by compatible probability,which is more intuitionistic and veracious. It solves the problem of uncertain samples,and lowers the error rate of classification.Finally,the article proposes the classification based on the covering generalized decision rule of evidence theory .It proposes three methods based on evidence theory and covering rough set, denoted as method one,method two and method three.It also proposes four methods based on evidence theory and covering transformation, denoted as method a,method b,method c and method d. Comparing with all kinds of classification methods,the article proves that the error rate obtained by the covering transformation is lower than that of covering rough set.