An anonymization method combining anatomy and permutation for protecting privacy in microdata with multiple sensitive attributes

Abstract

Microdata, such as customer transactional data, play an important role in data mining area. To protect privacy of individuals, microdata should be anonymized or desensitized before publishing or mining it. Anatomy is a popular technique to anonymize microdata. However, the Anatomy technique cannot resist linking attacks for the reason that it does not distort quasi-identifier attributes. To overcome the disadvantage of anatomy, we propose a method combining anatomy with permutation to anonymize microdata. The proposed method anonymizes microdata in two major steps: (1) anatomizing microdata, and (2) permutating quasiidentifier attributes. To realize the anonymization method, we further propose two algorithms, namely the NMBPA (Naive Multi-sensitive Bucketization Permutation Algorithm) and the CDMBPA (Closest Distance Multi-sensitive Bucketization Permutation Algorithm). Experimental results show that the proposed method can deal with linking attacks effectively, i.e., generate high quality anonymous data with low suppression ratios.