Research On Knowledge Change Rate-attributes Importance Measure Methods

With the constantly updated of computer network technology, amount of information in the database is increasingly large. In order to select practical valuable data from massive database, data mining technology will be developed. Therefore, how to remove the non-critical data set in database(i.e. attributes reduction) is a key part in data mining technology.For the problem of attributes importance measure, in this paper, taking knowledge hiding in decision information system as a carrier, and knowledge change caused by attributes set change as a basis, we first discuss associated features among positive region,lower(or upper) approximation of decision classes and knowledge in system, then give several knowledge change rate-based attributes importance measure methods(BKCR-AIM for short) satisfying structural features of fuzzy measure. Secondly, we discuss structural features and constructed strategies of BKCR-AIM, then further analyze features of BKCR-AIM combining with a specific case. Finally, example calculation shows that it is a feasible way through knowledge change in system to consider attributes importance, and the established mode that comprehensive knowledge change rate-based attributes importance measure method(BCKCR-AIM for short) has good structural features and strong interpretability.At the same time, we propose a reduction algorithm based on comprehensive knowledge change rate(BCKCR-ARA for short), and further analyze its features,feasibility and effectiveness through combining database UCI. Finally, results show that BCKCR-ARA can not only effectively realize attributes reduction, but also has good interpretability and operability. Thus, by using the algorithm in practical problems,managers can make rational decisions in an uncertain environment. Then, we can effectively utilize existing information systems to summarize associated features among attributes. And it also has broad application prospects in knowledge acquisition,information integration, comprehensive evaluation, fuzzy decision making, pattern recognition, comprehensive evaluation and other fields.