Progressive Subspace Skyline Clusters Mining on High Dimensional Data

Abstract

Skyline queries have caused much attention for it helps users make intelligent decisions over complex data. Unfortunately, too many or too few skyline objects are not desirable for users to choose. Practically, users may be interested in the skylines in the subspaces of numerous candidate attributes. In this paper, we address the important problem of recommending skyline objects as well as their neighbors in the arbitrary subspaces of high dimensional space. We define a new concept, subspace skyline cluster, which is a compact and meaningful structure to combine the advantages of skyline computation and data mining. Two algorithms Sorted-based Subspace Skyline Clusters Mining, and Threshold-based Subspace Skyline Clusters Mining are developed to progressively identify the skyline clusters. Our experiments show that our proposed approaches are both efficient and effective.