A Practical Anonymization Approach for Imbalanced Datasets

초록

Person-specific data owned by different data holders is usually anonymized before being shared with researchers or data miners. Anonymization is a pertinent solution for releasing useful information while ensuring privacy. Many anonymization approaches have been proposed but majority of the existing approaches do not consider the influence of user's attributes on privacy and utility. Consequently, privacy preservation and utility enhancement become challenging and particularly difficult when anonymizing imbalanced datasets that contain less heterogeneous values. To address these problems for imbalanced datasets, we propose a practical anonymization approach that effectively preserves users' privacy while maintaining high utility of anonymous data. It quantifies the influence of attributes on the degree of user's reidentification in order to protect user's privacy. Data transformation is performed adjustably considering the influence of users' attributes and their distributions. Experimental results obtained from real-world datasets show the efficacy of our approach and verify the abovementioned assertions.

키워드