Privacy preserving DBSCAN for vertically partitioned data

Abstract

Clustering algorithms are attractive for the core task of class identification in large databases. In recent years privacy issues also became important for data mining. In this paper, we construct a privacy preserving version of the popular clustering algorithm DBSCAN. This algorithm is density-based. Such notion of clustering allows us to discover clusters of arbitrary shape. DBSCAN requires only two input parameters, but it offers some support in determining appropriate values. Originally, DBSCAN uses R-Trees to support efficient associative queries. Thus, one solution for privacy preserving DBSCAN requires to have privacy preserving R-Trees. We achieve this here. © Springer-Verlag Berlin Heidelberg 2006.