Fast algorithms for mining emerging patterns

Abstract

Emerging Patterns are itemsets whose supports change significantly from one dataset to another. They are useful as a means of discovering distinctions inherently present amongst a collection of datasets and have been shown to be a powerful technique for constructing accurate classifiers. The task of finding such patterns is challenging though, and efficient techniques for their mining are needed. In this paper, we present a new mining method for a particular type of emerging pattern known as a jumping emerging pattern. The basis of our algorithm is the construction of trees, whose structure specifically targets the likely distribution of emerging patterns. The mining performance is typically around 5 times faster than earlier approaches. We then examine the problem of computing a useful subset of the possible emerging patterns. We show that such patterns can be mined even more efficiently (typically around 10 times faster), with little loss of precision. © 2002 Springer-Verlag Berlin Heidelberg.