Modeling & Predicting Privacy Risk in Personal Data

Privacy in Big Data analytics is one of the most important issues that analysts and businesses face when managing personal data. In a privacy preserving analysis process, the privacy risk on the individuals represented in the data is firstly evaluated, then the data is appropriately modified in order to preserve privacy while at the same time maintaining a certain level of data quality. In this thesis we focus on privacy risk assessment, proposing new models and algorithms to deal with this fundamental part of privacy aware systems. We propose some extensions to an existing state-of-the-art privacy risk assessment framework, to improve on existing literature. Then, we propose a classification based methodology to predict privacy risk. We validate our proposal on three different types of real world data: human mobility, retail and social network data. Finally we propose a new model for the behavior of an adversary in human mobility data, leveraging the natural structure and constraints of this kind of data.