ONN the use of neural networks for data privacy

Abstract

The need for data privacy motivates the development of new methods that allow to protect data minimizing the disclosure risk without losing valuable statistical information. In this paper, we propose a new protection method for numerical data called Ordered Neural Networks (ONN). ONN presents a new way to protect data based on the use of Artificial Neural Networks (ANNs). The main contribution of ONN is a new strategy for preprocessing data so that the ANNs are not capable of accurately learning the original data set. Using the results obtained by the ANNs, ONN generates a new data set similar to the original one without disclosing the real sensible values. We compare our method to the best methods presented in the literature, using data provided by the US Census Bureau. Our experiments show that ONN outperforms the previous methods proposed in the literature, proving that the use of ANNs is convenient to protect the data efficiently without losing the statistical properties of the set. © Springer-Verlag Berlin Heidelberg 2008.