A novel approach for mining emerging patterns in rare-class datasets

Abstract

Mining emerging patterns (EPs) in rare-class databases is one of the new and difficult problems in knowledge discovery in databases (KDD). The main challenge in this task is the limited number of rareclass instances. This scarcity limits the number of emerging patterns that can be mined for the rare class. In this paper, we propose a novel approach for mining emerging patterns in rare-class datasets. We experimentally prove that our method is capable of gaining enough knowledge from the rare class; hence, it increases the performance of EPbased classifiers. © 2007 Springer.