A lossless representation for association rules satisfying multiple evaluation criteria

Abstract

A lot of data mining literature is devoted to association rules and their evaluation. As the number of discovered rules is often huge, their direct usage by a human being may be infeasible. In the case of classical strong association rules, which are defined as rules supported by sufficiently large fraction of data and having sufficient confidence, a number of their concise representations have been proposed. However, as indicated in the literature, support and confidence measures of association rules seem not to cover many aspects that could be of interest to a user. In consequence, many other measures have been proposed to evaluate association rules. In this paper, we identify an important and wide class of rule ACBC-evaluation measures and offer a lossless representation of association rules satisfying constraints for any set of evaluation measures from this class. A number of properties of the representation is derived as well.