Effective algorithm for discovering fuzzy rules in relational databases

In this paper, we present a novel technique, called F-APACS, for discovering fuzzy association rules in relational databases. Instead of dividing up quantitative attributes into fixed intervals and searching for rules expressed in terms of them, F-APACS employs linguistic terms to represent the revealed regularities and exceptions. The definitions of these linguistic terms are based on fuzzy set theory and the association rules expressed in them are, therefore, called fuzzy association rules here. To discover these rules, F-APACS utilizes an objective interestingness measure when determining if two attribute values are related. This measure is defined in terms of an `adjusted difference' between observed and expected frequency counts. The use of such a measure has the advantage that no user-supplied thresholds are required. In addition to this interestingness measure, F-APACS has another unique feature that it provides a mechanism to allow quantitative values be inferred from the rules. Such feature, as shown here, make F-APACS very effective at various mining tasks.