抽象的

Detection and Prevention of Leaks in Anonymized Datasets

Sandeep Varma Nadimpalli and Valli Kumari Vatsavayi

With a wide spread of modern technology, person specificdata dissemination has beenincreasing rapidly,leading to a global concern for preserving privacy of an individual. Several principles like k-anonymity, l-diversity etc., have been proposed to protect the person specific information during data publishing. However, the presence of dependencies in an anonymized dataset may identify the individual due to the hypothetical nature of the adversary/attacker. This paper shows how the presence of these dependencies among Quasi-Identifiers (QI), Sensitive (S) attributes and also between QI and S attributes can lead to the potential identification of an individual using Bayesian Networks. A solution Break-Merge (BM) was proposed on the fly to reduce the attacker?s inferring nature on the sensitive data. Experimentations show the efficacy of theproposed approaches.