Mining fuzzy association rules

Abstract

In this paper, we introduce a novel technique, called F-APACS, for mining fuzzy association rules. Existing algorithms involve discretizing the domains of quantitative attributes into intervals so as to discover quantitative association rules. These intervals may not be concise and meaningful enough for human experts to easily obtain nontrivial knowledge from those rules discovered. Instead of using intervals, F-APACS employs linguistic terms to represent the revealed regularities and exceptions. The linguistic representation is especially useful when those rules discovered are presented to human experts for examination. The definition of linguistic terms is based on fuzzy set theory and hence we call the rules having these terms fuzzy association rules. The use of fuzzy techniques makes F-APACS resilient to noises such as inaccuracies in physical measurements of real-life entities and missing values in the databases. Furthermore, F-APACS employs adjusted difference analysis which has the advantage that it does not require any user-supplied thresholds which are often hard to determine. The fact that F-APACS is able to mine fuzzy association rules which utilize linguistic representation and that it uses an objective yet meaningful confidence measure to determine the interestingness of a rule makes it very effective at the discovery of rules from a real-life transactional database of a PBX system provided by a telecommunication corporation.