Shapley Values Enhance Data Masking for Privacy and Utility

The research introduces a new framework that applies Shapley-value-based feature attribution to the domain of data privacy, specifically for data masking. This approach aims to address the critical balance between minimizing disclosure risk and preserving data utility. Unlike existing literature that often focuses on this trade-off at the dataset level, this framework tackles it at the individual feature level. The proposed method is designed to be agnostic to the specific data masking techniques, statistical or machine learning methods, and evaluation metrics for utility and risk. Experimental results indicate that this Shapley-value-based framework effectively reduces disclosure risk while maintaining data utility. By providing a fair feature attribution, it offers a more granular and adaptable approach to privacy-preserving data transformations.