Km-anonymity for continuous data using dynamic hierarchies

Abstract

Many organizations, enterprises or public services collect and manage personal data of individuals. These data contain knowledge that is of substantial value for scientists and market experts, but carelessly disseminating them can lead to significant privacy breaches, as they might reveal financial, medical or other personal information. Several anonymization methods have been proposed to allow the privacy preserving sharing of datasets with personal information. Anonymization techniques provide a trade-off between the strength of the privacy guarantee and the quality of the anonymized dataset. In this work we focus on the anonymization of sets of values from continuous domains, e.g., numerical data, and we provide a method for protecting the anonymized data from attacks against identity disclosure. The main novelty of our approach is that instead of using a fixed, given generalization hierarchy, we let the anonymization algorithm decide how different values will be generalized. The benefit of our approach is twofold: a) we are able to generalize datasets without requiring an expert to define the hierarchy and b) we limit the information loss, since the proposed algorithm is able to limit the scope of the generalization. We provide a series of experiments that demonstrate the gains in terms of information quality of our algorithm compared to the state-of-the-art.