Representative Rule Templates for Association Rules Satisfying Multiple Canonical Evaluation Criteria

Originally, strong association rules were defined as those that have sufficiently high values of two parameters: support and confidence. However, it has been shown in the literature that in general neither of these two measures is capable of determining (in)dependence between rules’ constituents correctly. Thus, usage of other measures for rule evaluation is also under research. In this paper, we formulate a generic notion of a canonical measure and show important examples of canonical measures. For association rules satisfying any set of criteria based on canonical measures, we offer their concise lossless representation in the form of so called representative rule templates. Also, we derive a number of properties of this representation.