Exploratory Quantitative Contrast Set Mining: A Discretization Approach

Contrast sets have been shown to be a useful tool for describing differences between groups. A contrast set is a set of association rules for which the antecedents describe distinct groups, a common consequent is shared by all the rules, and support for the rules is significantly different between groups. While techniques for generating contrast sets containing categorical attributes in the consequent are "straightforward", techniques for generating contrast sets containing continuous-valued attributes are not. In this paper, we describe a technique for generating contrast sets describing the differences between two groups, where the consequent in the rules contains up to two continuous-valued attributes. We propose a modified equal- width binning interval approach to discretizing continuous-valued attributes, where the approximate width of the desired intervals is provided as a parameter to the model. We also propose an objective measure for identifying and ranking the potentially interesting contrast sets. Experimental results demonstrate the effectiveness of our approach and the utility of the interest measure.