A New Mathematical Optimization-Based Method for the m-invariance Problem

TL;DR

This paper introduces a novel mathematical optimization heuristic based on column generation to improve solutions for the NP-hard m-invariance and { au}-safety problems, significantly reducing information loss in dynamic data privacy.

Contribution

It presents a new optimization-based heuristic for the NP-hard m-invariance and { au}-safety problems, outperforming existing heuristics in data privacy utility.

Findings

Achieved solutions with ILs as low as 1.87, 8.5, and 1.93.

Outperformed current heuristics with ILs of 39.03, 51.84, and 57.97.

Demonstrated significant utility improvements in dynamic data privacy.

Abstract

The issue of ensuring privacy for users who share their personal information has been a growing priority in a business and scientific environment where the use of different types of data and the laws that protect it have increased in tandem. Different technologies have been widely developed for static publications, i.e., where the information is published only once, such as k-anonymity and {\epsilon}-differential privacy. In the case where microdata information is published dynamically, although established notions such as m-invariance and {\tau}-safety already exist, developments for improving utility remain superficial. We propose a new heuristic approach for the NP-hard combinatorial problem of m-invariance and {\tau}-safety, which is based on a mathematical optimization column generation scheme. The quality of a solution to m-invariance and {\tau}-safety can be measured by the Information Loss (IL), a value in [0,100], the closer to 0 the better. We show that our approach improves by far current heuristics, providing in some instances solutions with ILs of 1.87, 8.5 and 1.93, while the state-of-the art methods reported ILs of 39.03, 51.84 and 57.97, respectively.