Research On Data Perturbation Privacy Preserving Method For Distributed Clustering

In recent years, the parties and organizations of various fields stored huge volume of data and begin to apply distributed clustering methods on those data for data sharing in order to gain much more benefits. However, privacy information must be protected during the process of data sharing. Absolutely, how to improve the performance of the distributed privacy preserving clustering methods has already became a new focus.After an intensive study on the exsiting privacy preserving distort methods, it can be concluded that most of the methods try to find a balance between the privacy degree and data utility. In addation, almost all the datasets that various parties hold is high-dimension and have huge sizes. As a result, efficiency should also be considered when designing the method otherwise the processing time will be intolerable. To meet these needs, a new data perturbation method is proposed in this thesis called the Centro Symmetry Based Transformation--CST. The method first generates a symmetry center then distorts the data set by centro symmetry transformation with the symmetry center. There are three main steps of the algorithm, including determing the probability distribution parameters and the range of the symmetric center, generation of the symmetric center and the disturbance of the data sets. The CST method improves the privacy degree by a certain extent while ensures the utility of the perturbed data and can resist certain type of malicious attack as well.In response to the participated organizations steal private information by collusion, this paper proposes an improved method of CST method, by sacrificing a certain degree of data utility to improve the algorithm's resistance to attack.At last, a series of experiments demonstrate that this method is feasible and efficient, and the dirction of the futher exploration is discussed.