Synthetic Data Satisfying the Requirements of a New Attribute Disclosure Risk Criterion

• Activity Number: 423
• Type: Contributed
• Date/Time: Tuesday, August 11, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Government Statistics Section
• Author(s): Anna Oganyan*
• Companies: National Center for Health Statistics
• Keywords: synthetic data ; attribute disclosure ; mixture model ; constraints ; cluster volumes ; EM algorithm

Abstract

In this paper we present an approach for the generation of synthetic microdata that satisfies the requirements of a new attribute disclosure risk criterion. First, we define a metric of attribute disclosure which is called v-dispersion. This metric quantifies the risk based on the spread of the multidimensional confidence regions for the original data values. Next we describe a method that satisfies the requirements of v-dispersion. This method is based on a mixture model with constraints on parameters of components' spread. Experiments with real data show that the proposed approach compares very favorably with other methods of disclosure limitation in terms of utility and risk.