SkyDist: Data Mining on Skyline Objects

The skyline operator is a well established database primitive which is traditionally applied in a way that only a single skyline is computed. In this paper we use multiple skylines themselves as objects for data exploration and data mining. We define a novel similarity measure for comparing different skylines, called SkyDist. SkyDist can be used for complex analysis tasks such as clustering, classification, outlier detection, etc. We propose two different algorithms for computing SkyDist, based on Monte-Carlo sampling and on the plane sweep paradigm. In an extensive experimental evaluation, we demonstrate the efficiency and usefulness of SkyDist for a number of applications and data mining methods.