How to enable fine-grained mobile network operator data analytics: The power of perturbative statistical disclosure control

Abstract

Mobile network operator (MNO) data provide rich, high-resolution information that is valuable for a wide range of statistical analyses. In Germany, however, current data-privacy practices severely restrict the release of high-resolution datasets. This paper demonstrates that perturbative statistical disclosure control (SDC) methods can increase the utility of MNO data without compromising privacy by enabling the publication of higher-resolution aggregates under controlled noise mechanisms. After a brief overview of MNO data and the prevailing German privacy regulations, I introduce two tabular SDC techniques — the Cell-Key Method (CKM) and the Laplace Mechanism as an implementation of the Differential Privacy (DP) framework — and discuss their suitability for mobility data. I conduct an experimental evaluation using a dataset with device movements derived from real MNO records, generated at both low and high temporal resolutions. A systematic grid search over the space of the most important parameters of CKM and DP is performed, with the focus on utility rather than a full privacy-risk assessment. Results show that CKM delivers strong utility, especially for attribute combinations with low case counts, while DP either introduces unrealistic negative numbers or markedly inflates device movements. Compared with the traditional k-anonymity approach, perturbative SDC achieves comparable utility for k ≥ 5 and additionally permits the release of aggregates below this threshold, supporting finer-grained analyses. CKM also offers advantages in stakeholder communication and ease of implementation, although its integration into distributed computing environments remains an open research question. Overall, CKM emerges as a promising candidate for securely exporting high-resolution, aggregated MNO data to national statistical institutes, providing improved utility compared to the traditional k-anonymity approach while maintaining robust privacy protection.