Approximation algorithms for optimizing privacy and utility

Abstract

There are two major aspects of Data Privacy –identity disclosure and attribute disclosure. A number of algorithms have been developed to protect identity disclosure. However, many of these algorithms have not been much successful in preventing attribute disclosure. The standard implementation techniques for the commonly used privacy models are NP-hard. They are time-consuming and often unnecessary. This paper investigates the existing algorithms for preserving privacy by preventing identity disclosure. Improved approximation algorithms are proposed to prevent identity disclosure and attribute disclosure. The existing approximation algorithm for k-anonymity is extended by applying simulated annealing to reduce attribute disclosure. The concepts of l-diversity and closeness are used for minimizing attribute disclosure in polynomial time. An improved algorithm for calculating the value of t-closeness is also proposed. A new parameter is introduced to anonymize the dataset considering both privacy and utility. An existing algorithm has been modified to consider the utility while determining the best full domain generalization scheme. This is followed by a proposed approximation algorithm which takes into consideration the utility while preserving the privacy. It is an example of partial domain generalization wherein different tuples can be anonymized to different extents.