View Related Documents

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.

Fulltext Preview

Image of the first page of the fulltext document