Privacy-Preserving Data Mining Based on Sample Selection and Singular Value Decomposition
Guang Li, Yadong Wang · 2011
For improving the PPDM (privacy-preserving data mining) methods based on matrix decomposition, this paper proposed a new PPDM method both using sample selection and matrix decomposition. The original matrix decomposition-based methods perform attribute extraction by matrix decomposition to analyze data, find the important information for data mining and remove the unimportant information to perturb data. In addition to attribute extraction, sample selection also can analyze data. If both sample selection and matrix decompositions are used, the important information for data mining should be found more accurately, which is the basic idea of this proposed new method. The experiments showed that this new method can perform better in privacy preserving than the methods using matrix decompositions alone, while keeping data utility.