期刊名称：International Journal of Hybrid Information Technology
印刷版ISSN：1738-9968
出版年度：2016
卷号：9
期号：7
页码：79-88
DOI：10.14257/ijhit.2016.9.7.09
出版社：SERSC
摘要：Aiming at the incomplete information systems on the condition of no prior domainknowledge, several known model extension based on the rough set theory are introduced atfirst, such as the tolerance relation, non-symmetric similarity relation, limited tolerancerelation and valued tolerance relation. Then the merits and drawbacks of several existingvalued tolerance relations are compared in this article. Next, the experiments on some UCIdata sets have been done ,based on the experimental result, the author discuss therelationship between the threshold selection and classification accuracy of statistical valuedtolerance relation(SVT) . Directing at the difficulty of selecting a suitable threshold, theauthor presents a new improved valued tolerance relation (NVT) which can choose properthreshold automatically on the basis of each data set’s feature. Experiment results indicatethat the new relation can get better classification accuracy than the other extension models indealing with the incomplete system which has small incomplete degree
关键词：Incomplete information system; rough set; New Valued Tolerance relation;(NVT); statistical valued tolerance relation (SVT); threshold selection
