DP-AP: Differential Privacy-Preserving Affinity Propagation Clustering

The affinity propagation (AP) clustering can accurately and effectively process complex data and has been widely used in various data clustering analysis fields, especially in computer vision and computational biology. However, the dataset usually contains sensitive information. To prevent the leakage of users' privacy during the clustering process, this paper proposes an affinity propagation clustering with differential privacy algorithm. Based on the affinity propagation clustering, we first adjust the corresponding preference value by calculating the density value of each sample and then add Laplace noise to the responsibility matrix to protect the data privacy. Experimental results show that the adjusted preference value with the sample density can effectively reduce the running time of the algorithm and the loss of data accuracy caused by differential privacy.