Representation by levels: an alternative to fuzzy sets for fuzzy data mining

Abstract

In this paper we describe and discuss the main contributions of the representation by levels approach to fuzzy data mining. Representation by levels is an alternative representation of fuzziness in information and data, which is complementary to fuzzy sets in the sense that it provides tools and algebraic structures beyond the capabilities of fuzzy set theories, based on t-norms, t-conorms and fuzzy negations. Our approach allows to extend any crisp mining technique to the fuzzy case in a simple way, keeping all the properties of the crisp technique. We illustrate our discussion with examples and existing approaches based on representation by levels to fuzzy association rules and the related issues of mining exception/anomalous rules and mining fuzzy bag databases.