An anti-collusion solution for privacy-preserving data mining

Abstract

In distributed data mining with privacy preserving, the algorithms which adopt data obscurity method are sometimes vulnerable facing collusion. In this research, such stream of collusion challenge is defined. Since previously suggested methods would cause multiple orders magnitude of communication or partial security covering among parties, we have proposed a method (Adjacency Permutation) for the colluding problem and integrated it in establishment of RPA (Ring Polling for Association rules) platform. The proposed solution is (1) lightweight for only increasing linear communication; (2) effective for all sites are covered by suggested protection. In synthetic simulation, RPA is compared with other two representative DDM (distributed data mining) algorithms (FDM and CER); the results show good performance of RPA on efficiency and effectivity.