Computing Exact Skyline Probabilities for Uncertain Databases

Abstract—With the rapid increase in the amount of uncertain data available, probabilistic skyline computation on uncertain databases has become an important research topic. Previous work on probabilistic skyline computation, however, only identifies those objects whose skyline probabilities are higher than a given threshold, or is useful only for two-dimensional data sets. In this paper, we develop a probabilistic skyline algorithm called PSkyline which computes exact skyline probabilities of all objects in a given uncertain data set. PSkyline aims to identify blocks of instances with skyline probability zero, and more importantly, to find incomparable groups of instances and dispense with unnecessary dominance tests altogether. To increase the chance of finding such blocks and groups of instances, PSkyline uses a new in-memory tree structure called Z-tree. We also develop an online probabilistic skyline algorithm called O-PSkyline for uncertain data streams and a top-k probabilistic skyline algorithm called K-PSkyline to find top-k objects with the highest skyline probabilities. Experimental results show that all the proposed algorithms scale well to large and high-dimensional uncertain databases. Index Terms—Skyline computation, skyline probability, uncertain database, data stream. F