A quantification of cluster novelty with an application to Martian topography

Abstract

Automated tools for knowledge discovery are frequently invoked in databases where objects already group into some known classification scheme. In the context of unsupervised learning or clustering, such tools delve inside large databases looking for alternative classification schemes that are both meaningful and novel. A quantification of cluster novelty can be looked upon as the degree of separation between each new cluster and its most similar class. Our approach models each cluster and class as a Gaussian distribution and estimates the degree of overlap between both distributions by measuring their intersecting area. Unlike other metrics, our method quantifies the novelty of each cluster individually, and enables us to rank classes according to its similarity to each new cluster. We test our algorithm on Martian landscapes using a set of known classes called geological units; experimental results show a new interpretation for the characterization of Martian landscapes. © Springer-Verlag Berlin Heidelberg 2004.