Sachiko Suzuki

A Distance-Based Clustering and Selection of Association Rules on Numeric Attributes

Association rule is a kind of important knowledge extracted from databases. However, a large number of association rules may be extracted. It is difficult for a user to understand them. How to select some “representative” rules is thus an important and interesting topic. In this paper, we proposed a distance-based approach as a post-processing for association rules on numeric attributes. Our approach consists of two phases. First, a heuristic algorithm is used to cluster rules based on a matrix of which element is the distance of two rules. Second, after clustering, we select a representative rule for each cluster based on an objective measure. We applied our approach to a real database. As the result, three representative rules are selected, instead of more than 300 original association rules.