Undirected exception rule discovery as local pattern detection

Abstract

In this paper, we give an interpretation of our undirected exception rule discovery as local pattern detection and introduce some of our endeavors. Our undirected exception rule discovery outputs a set of rule pairs, each of which represents a pair of strong rule and its exception rule. A local pattern is defined as a pattern which deviates from a global model, and can be considered to correspond to our exception rule if the global model corresponds to our strong rule. Several attempts for undirected exception rule discovery are introduced in the context of local pattern detection. Our results mainly concern interestingness measure, algorithmic issues, noise modeling, and performance evaluation. © Springer-Verlag Berlin Heidelberg 2005.