Sensitivity Based Anonymization with Multi-dimensional Mixed Generalization

Abstract

Sensitive information about individuals must not be revealed when sharing data, but a data set must remain useful for research and analysis when published. Anonymization methods have been considered as a possible solution for protecting the privacy of individuals. This is achieved by transforming data in a way that guarantees a certain degree of protection from re-identification threats. In the process, it is important to ensure that the quality of data is preserved. K-anonymity is the most commonly used approach for the anonymization of published datasets. However, the approach causes a decline in data utility. The key challenge for data publishers is how to anonymize data without causing a significant decline in data utility. The paper addresses this challenge by proposing a multidimensional mixed generalization. We conduct experiments with mixed generalization. Our results show that mixed generalization preserves the quality of data for classification.