Mining implication-based fuzzy association rules in databases

Abstract

This chapter introduces an alternative type of fuzzy association rule, which is logic-oriented and makes use of a multiple-valued implication operator in order to connect the condition and conclusion parts of a rule. Consequently, the two parts do no longer play symmetric roles in associations, a property that appears quite reasonable and that avoids some difficulties of the classical (conjunction-based) approach. Among other aspects, fuzzy sets avoid an arbitrary determination of crisp boundaries for intervals. For this type of association, adequate quality measures are proposed and some semantic issues are discussed. In the field of data mining, so-called association rules (or associations for short) have gained considerable attraction. Such rules, syntactically written as A⇀B, provide a means for representing dependencies between attributes in a database. A and B denote sets of binary attributes, also called features or items. The intended meaning of a (binary) rule A⇀B is that a data record (or transaction) stored in the database that contains the set of items A is likely to contain the items B as well. Several efficient algorithms have been devised for mining association rules in large databases. Moreover, fuzzy association rules are very interesting from a knowledge representational point of view.