Credibility coefficients based on frequent sets

Abstract

Credibility coefficients are heuristic measures applied to objects of information system. Credibility coefficients were introduced to assess similarity of objects in respect to other data in information systems or decision tables. By applying knowledge discovery methods it is possible to gain some rules and dependencies between data. However the knowledge obtained from the data can be corrupted or incomplete due to improper data. Hence identification of these exceptions cannot be overestimated. It is assumed that majority of data is correct and only a minor part may be improper. Credibility coefficients of objects should indicate to which group a particular object probably belongs. A main focus of the paper is set on an algorithm of calculating credibility coefficients. This algorithm is based on frequent sets, which are produced while using data analysis based on the rough set theory. Some information on the rough set theory is supplied to enable expression of credibility coefficient formulas. Implementation and applications of credibility coefficients are presented in the paper. Discussion of some practical results of identifying improper data by credibility coefficients is inserted as well.