K-Anonymity through the Enhanced Clustering Method

Abstract

With the rise of the Social Web, there is increasingly more tendency to share personal records, and even make them publicly available on the Internet. However, such a wide spread disclosure of personal data has raised serious privacy concerns. If the released dataset is not properly anonymized, individual privacy will be at great risk. K-anonymity is a popular and practical approach to anonymize datasets. In this study, we use a new clustering approach to achieve k-anonymity through enhanced data distortion that assures minimal information loss. During a clustering process, we include an additional constraint, minimal information loss, which is not incorporated into traditional clustering approaches. Our proposed algorithm supports a data release process such that data will not be distorted more than they are needed to achieve k-anonymity. We also develop more appropriate metrics for measuring the quality of generalization. The new metrics are suitable for both numeric and categorical attributes. Our experimental results show that the proposed algorithm causes significantly less information loss than existing clustering algorithms.