SDkA: Synthetic Data Integrated k-Anonymity Model for Data Sharing With Improved Utility

초록

This paper presents a novel anonymity model named "SDkA" that establishes the synergy between synthetic data (SD) and a k-anonymity (k-A) model as a privacy-utility enhancer in data sharing scenarios. SDkA introduces a data-level solution for data quality, diversity, and quantity enrichment and controls higher modifications in the data. Specifically, SD produced by a conditional generative adversarial network, and an automated method for fixing commonly encountered quality-related problems, are introduced to address size, quantity, and diversity issues. By introducing these enhancements in the data, a better defense against diverse privacy attacks is accomplished, compared to a standard k-A model. Following this, the selective generalization concept is introduced as an optimization in the conventional k-A model to prevent unnecessary generalization, thereby further fortifying the k-A model to yield enhanced utility. Detailed experiments prove that SDkA effectively defends against diverse types of attacks and significantly improves the utility of the anonymized data, compared to the k-A model.