Impact of outliers on anonymized categorical data

Abstract

Preserving privacy is indispensable when publishing microdata with sensitive information. Anonymization principles like k-anonymity, l-diversity were developed to protect the sensitive information. An adversary with sufficient background knowledge inferring the individual's sensitive information signifies disclosure of the microdata. None of the above mentioned principles addressed the presence of outliers. Outliers can be classified into two types viz., local and global. This paper proposes a practically feasible distance based algorithm to anonymize the local outliers. Our proposed algorithm is capable of handling both numerical and categorical data. The experimental results of our proposed approach focused to categorical data presented. © 2011 Springer-Verlag.