K-modes Based Categorical Data Clustering Algorithms Satisfying Differential Privacy

As one of the important algorithms in data mining, clustering algorithm has a wide range of applications in the real world. However, clustering algorithms have the risk of privacy leakage, such as the k-modes algorithm for clustering categorical data. Therefore, this paper proposes k-Modes based clustering algorithms for categorical data satisfying differential privacy. In the process of clustering iteration, the Laplace mechanism, Gaussian mechanism or Exponential mechanism is used to interfere the clustering algorithm to protect the privacy of data. And, in order to improve the accuracy of the clustering results, we changed the selection method of the initial center points. At the end of the paper, we compare the influence of different differential privacy mechanisms on the accuracy of clustering algorithm through experiments.