A Computational Evaluation of Optimization Solvers for CTA

Minimum-distance controlled tabular adjustment methods (CTA), an its variants, are considered an emerging perturbative approach for tabular data protection. Given a table to be protected, the purpose of CTA is to find the closest table that guarantees protection levels for the sensitive cells. We consider the most general CTA formulation which includes binary variables, thus providing protected tables with a higher data utility, at the expense of a larger solution time. The resulting model is a Mixed Integer Linear Problem (MILP). The purpose of this work is twofold. First, it presents and describes the main features of a package for CTA which is linked to both commercial (Cplex and Xpress) and open-source (Glpk, Cbc and Symphony ) MILP solvers. The particular design of the package allows easy integration with additional solvers. The second objective is to perform a computational evaluation of the above two commercial and three open-source MILP solvers for CTA, using both standard instances in the literature and real-world ones. Users of tabular data confidentiality techniques in National Statistical Agencies may find this information useful for the trade-off between the (more efficient but expensive) commercial and the (slower but free) open-source MILP solvers.