K-anonymity algorithm based on improved clustering

Abstract

K-anonymity is the most widely used technology in the field of privacy preservation. It has a good performance particularly in protecting data privacy in the scenarios of data publication, location-based service and social network. In this paper, we propose a new algorithm to achieve k-anonymity in a better way through improved clustering, and we optimize the clustering process by considering the overall distribution of quasi-identifier groups in a multidimensional space. With the local optimal clustering, we try our best to guarantee minimized intra-cluster distances and maximized inter-cluster distances. Therefore, the quality of anonymized data can be greatly improved. Compared with some popular algorithms like k-member, Mondrian, and one-time k-means, the experimental results show our algorithm can effectively reduce the information loss while generating equivalence classes. The total information loss of the anonymized dataset decreases by about 20% on average than that of other algorithms. It also performs well in dealing with both numerical attributes and categorical attributes.