Privacy Preserving Classification Algorithm Based Random Diffusion Map

In this paper, a privacy preserving classification algorithm based random diffusion map is presented. We first alter the selection of the parameter dimension d and metaparameter fixed value ¿ for satisfying the security of privacy-preserving classification. Further the sensitive attributes are embedded into random(even higher) dimension feature space using random diffusion map, thus the sensitive attributes are transformed and protected. Because the transformed space dimension d and the ¿ are both stochastic, this algorithm is not easily be breached. In addition, diffusion map can keep topology structure of dataset, so the classification precision after encryption are kept well. The experiment shows that the present method can provide sensitive information enough protect without much loss of the classification precision.