A Probabilistic Approach for On-Line Sum-Auditing

In this paper we consider the problem of auditing databases which support statistical sum-queries to protect the security of sensitive information. We study the special case in which the domain of the sensitive information is a discrete set; in particular, we focus on a boolean domain. Principles and techniques developed for the security of statistical databases in the case of continuous attributes do not apply here. We provide a probabilistic framework for the on-line sum-auditing and we show that sum-queries can be audited by means of a Bayesian network. Finally, we provide a preliminary analysis of the usefulness of the probabilistic approach.