A Data Privacy Preservation Method Based on Lossy Decomposition

Recently, privacy preserving data mining has been a hot topic in data mining research community. The data privacy preservation is one of the important issues of privacy preserving mining. Many methods have been presented for this problem. However, the existing methods often have the shortcomings of high loss distortion and less aggregate query accuracy on the private or anonymous dataset. In this paper, based on the existing (alpha, k) privacy preservation model, an improved method, Alpha+, is presented using the lossy decomposition theory of relation database. Alpha+ firstly use (alpha, k) method to generate the private dataset for the original database. Then the private dataset is projected into two separated tables NSS and SS. The two tables are related with each other through the same relation attributes and then the redundancy information of lossy join of them can be used to preserve the private information. Secondly, Alpha+ merges the same tuples of NSS and SS tables to reduce the size of them, and the modified NSS and SS are finally published. The security comparison analysis between Alpha+ and the other similar methods is also given. The experimental results show that Alpha+ outperforms the existing methods in terms of aggregate query accuracy on the private dataset.