Density Peaks Clustering with Differential Privacy

Density peaks clustering (DPC) is a latest and well-known density-based clustering algorithm which offers advantages for finding clusters of arbitrary shapes compared to others algorithm. However, the attacker can deduce sensitive points from the known point when the cluster centers and sizes are exactly released in the cluster analysis. To the best of our knowledge, this is the first time that privacy protection has been applied to DPC. In this paper, we provide density peaks clustering privacy protection(DPCP) model to obtain the clustering results without revealing the data via differential privacy protection, in which the privacy protection is achieved by add Laplace noise to local density ρ and distance δ. However, the computation complexity will reaches O(n) and have an inaccurate clustering results when adding noise to the data set directly. Therefore, we are inspired by the idea of divide and conquer algorithm. Firstly, we divide the data set into relatively independent groups by Voronoi diagram and then adding noises. We employ a parallel computing by MapReduce to improve the efficiency. Secondly, according to the principle that is the privacy budget can be superimposed in high dimensional data. We introduces 1+ 2-differential privacy protection model and ensure the accuracy of the calculation via data replication and filter. Where 1 and 2 to protect ρ and δ respectively. Finally, through a lot of experiments, we also provide performance analysis and privacy proof of our solution. CCS Concepts •Security and privacy → Domain-specific security and privacy architectures;