Optimizing Classification Accuracy with Simulated Annealing in k-Anonymity

In the era of extensive data collection, achieving a balance between individual privacy protection and the preservation of data utility is critical. This paper introduces a novel k-anonymization approach that integrates simulated annealing with generalization hierarchies and suppression constraints to optimize classification accuracy on anonymized datasets. Unlike traditional greedy algorithms, our method probabilistically navigates the anonymization solution space. We validate our approach through extensive experiments on two real-world datasets, Adult and MIMIC-III, comparing against the state-of-the-art ARX framework. Our method improves AUC-ROC scores by up to 3.3% over ARX, and successfully generates feasible anonymizations even under stringent privacy requirements where ARX fails – demonstrating robustness and effectiveness of our simulated annealing-based anonymization strategy.