Mining negative contrast sets from data with discrete attributes

摘要

Negative contrast sets show that the relationships between the existence of some characteristics and the nonexistence of some other characteristics for various groups are significantly different. These sets can provide additional information for making decisions. Mining positive contrast sets like ‘X and Y’ is relatively straightforward and computationally efficient with respect to mining negative contrast sets like ‘not X and Y’. In this paper, we propose an algorithm to discover negative contrasts sets across groups, and establish some properties to accelerate the execution of the algorithm. It is then applied on an insurance data set to find meaningful negative contrast sets that can assist in designing insurance programs for various types of customers.