Energy-efficient filtering for skyline queries in cluster-based sensor networks

Filtering is a generic technique for skyline retrieval in sensor networks, for the purpose of reducing the communication cost, the dominant part of energy consumption. The vast majority of existing filtering approaches are suitable for uniform and correlated datasets, whereas in many applications the data distribution is clustered or anti-correlated. The only work considering anti-correlated dataset requires significant energy for filtering construction, and it is hard to be efficiently adapted to clustered databases. In this paper, we propose a new filtering algorithm, which settles the problem by utilizing individual node characteristics and generating personalized filters. Given a fraction k, a personalized filter prunes at least k percent of points on assigned nodes. A novel scheme for data cluster representation and a sampling method are then proposed to reduce the filtering cost and maximize the benefit of filtering. Extensive simulation results show the superiority of our approach over existing techniques.