Dipartimento di Ingegneria informatica, automatica e gestionale

Health data anonymization is a hot topic, on which both the medical and the computer science communities have made a great effort to provide a safer and trustful way of sharing data among research centers and hospitals.The main challenge in data anonymization is to provide a proper trade off between the utility of the resulting data/models and protecting individual privacy.In this paper we present a real anonymization case, with particular emphasis on choices that have to be made to carry it on, and difficulties experienced using a data set with many dimensions, and not well distinguishable features. We present our approach for evaluating disclosure risks and methods for anonymising high-dimensional medical survey data and measuring the utility of the transformed data.