An Enhanced K-Anonymity Model against Homogeneity Attack

Abstract

k-anonymity is an important model in the field of privacy protection and it is an effective method to prevent privacy disclosure in micro-data release. However, it is ineffective for the attribute disclosure by the homogeneity attack. The existing models based on k-anonymity have solved this problem to a certain extent, but they did not distinguish the different values of the sensitive attribute, processed a series of unnecessary generalization and expanded the information loss when they protect the sensitive attribute. Based on k-anonymity, this paper proposed a model based on average leakage probability and probability difference of sensitive attribute value. It is not only an effective method to deal with the problem of attributes disclosure that k-anonymity cannot deal, but also to realize different levels of protection to the various sensitive attribute values. It has reduced the generalization to the data in the most possibility during the procedure and ensures the most effectiveness of quasi-identifier attributes. Greedy generalization algorithm based on the generalization information loss is also proposed in this paper. To choose the generalization attributes, the information loss is considered and the importance of generalization attribute to sensitive attribute is accounted as well. Comparison experiment and performance experiment are made to the proposed model. The experiment results show that the model is feasible.