Analysis of optimal privacy protection mechanism based on IT-MPD

Balancing privacy preservation and data utility remains a critical challenge in data-driven systems. Traditional metrics, such as k-anonymity and differential privacy, often adopt static thresholds or oversimplified assumptions, limiting their adaptability to dynamic privacy-utility trade-offs. This paper proposes Information-Theoretic Maximum Privacy Disclosure (IT-MPD), a novel metric leveraging entropy and KL divergence to quantify the worst-case privacy leakage. We formulate privacy protection as a constrained optimization problem, minimizing IT-MPD while ensuring mutual information between raw and anonymized data exceeds a utility threshold ϵ. By modeling the data release process as a Markov chain, we demonstrate that optimal mechanisms emerge at the Pareto frontier of privacy-utility trade-offs. Experiments on age datasets anonymized via K-anonymity show IT-MPD values decline from 6.644 (K=1) to 0.020 (K=50), while utility (measured by mutual information) drops from 4.605 to 0.693. Polynomial fitting of these trends enables practitioners to select K based on operational requirements - e.g., ϵ = 3 yields K=5 as the optimal parameter. This work bridges theoretical rigor with practical applicability, offering a dynamic framework for privacy compliance in heterogeneous data ecosystems.