Abstract
The use of Association Rule Mining techniques in diverse contexts and domainshas resulted in the creation of numerous interestingness measures. This, inturn, has motivated researchers to come up with various classification schemesfor these measures. One popular approach to classify the objective measures isto assess the set of mathematical properties they satisfy in order to helppractitioners select the right measure for a given problem. In this research,we discuss the insufficiency of the existing properties in literature tocapture certain behaviors of interestingness measures. This motivates us topresent a novel approach to analyze and classify measures. We refer to this asa rate of change analysis (RCA). In this analysis a measure is described by howit varies if there is a unit change in the frequency count$(f_{11},f_{10},f_{01},f_{00})$, for different pre-existing states of thefrequency counts. More formally, we look at the first partial derivative of themeasure with respect to the various frequency count variables. We then use thisanalysis to define two new properties, Unit-Null Asymptotic Invariance (UNAI)and Unit-Null Zero Rate (UNZR). UNAI looks at the asymptotic effect of addingfrequency patterns, while UNZR looks at the initial effect of adding frequencypatterns when they do not pre-exist in the dataset. We present a comprehensiveanalysis of 50 interestingness measures and classify them in accordance withthe two properties. We also present empirical studies, involving both syntheticand real-world datasets, which are used to cluster various measures accordingto the rule ranking patterns of the measures. The study concludes with theobservation that classification of measures using the empirical clusters sharesignificant similarities to the classification of measures done through theproperties presented in this research.