Desemantization for Numerical Microdata Anonymization

Many situations demand for publishing confidential data without revealing the identity of data owners. For this purpose, anonymization methods are specially important in order to minimize both the disclosure risk and the information loss of the released data.In this chapter, we describe a methodology for numerical data anonymization based on a novel strategy for preprocessing data. The key point of this strategy is to desemantize the data set, i.e. to gather all the values together in a single vector regardless of the attribute they belong to. Data anonymization is achieved by modeling the preprocessed data in a way that, it; is accurate enough to be representative of the original data, but sufficiently dissimilar not to reveal the confidential information.In order to prove the validity of our methodology, we present four different approaches for the modeling step of the anonymization process. Those approaches outperform) some of the best methods available in the literature in terms of the trade-off between privacy preservation and information loss.