Effective Privacy Preserved Clustering Based on Voronoi Diagram

Abstract

Consider a scenario like this: a data holder, such as a hospital (data publisher) wants to share patients' data with researcher (data user). However, due to privacy issue, the hospital could not publish the exact original data while the published data need to retain as much as possible the correlation of the original data for utility consideration. The entire existing models for publishing private data could not perfectly resolve the tradeoff between privacy and utility of the private data. This paper presents a novel private information publishing model Semi-Delaunay Diagram (SDD) based on Voronoi diagram and gives a clustering algorithm VDC based on SDD. This model not only protects privacy but also achieves a perfect clustering correlation. Extensive experiments show the different clustering results with the different input area parameter, and confirm that our VDC algorithm discovers clusters with arbitrary shape as DBSCAN algorithm does.