Inferential Disclosure Limitation in Multivariate Categorical Databases
Date of Award
Doctor of Philosophy in Computing Technology in Education (DCTE)
Graduate School of Computer and Information Sciences
Michael J. Laszlo
Protecting data against inferential disclosure is a significant research challenge. With the increasing pervasiveness of data warehouses and On-Line Analytical Processing (OLAP) applications, disclosure limitation in multidimensional databases is especially important. Recent research has proposed efficient methods for inferential disclosure detection in multidimensional categorical databases but has not addressed the problem of disclosure elimination. Disclosures are removed by additive noise data perturbation. The goal is to minimize information loss due to data perturbation. This dissertation formulates the disclosure elimination problem in multidimensional categorical databases as a constrained optimization model. Since finding optimal solutions to the resulting problem is computationally hard, a genetic algorithm (GA) is used to identify good feasible solutions. A greedy algorithm is developed to solve an important special case of the problem that involves non-zero disclosures. Results indicate that the proposed GA based approach and the greedy algorithm can efficiently identify good feasible solutions that require low levels of data perturbation. Extensive computational experiments are performed to validate these results.
Randy Lamar Justice. 2003. Inferential Disclosure Limitation in Multivariate Categorical Databases. Doctoral dissertation. Nova Southeastern University. Retrieved from NSUWorks, Graduate School of Computer and Information Sciences. (622)