Statistical Database Modeling for Privacy Preserving Database Generation

Abstract

Testing of database applications is of great importance. Although various studies have been conducted to investigate testing techniques for database design, relatively few efforts have been made to explicitly address the testing of database applications which requires a large amount of representative data available. As testing over live production databases is often infeasible in many situations due to the high risks of disclosure of confidential information or incorrect updating of real data, in this paper we investigate the problem of generating synthetic database based on a-priori knowledge about production database. Our approach is to fit general location model using various characteristics (e.g., constraints, statistics, rules) extracted from production database and then generate synthetic data using model learnt. As characteristics extracted may contain information which may be used by attacker to derive some confidential information, we present a disclosure analysis method which is based on cell suppression technique. Our method is effective and efficient to remove aggregate private information during data generation.

This research was supported by USA National Science Foundation Grant CCR-0310974.