Research Of Density Peak Clustering Algorithm Based On Differential Privacy Preserving

Data mining is a technique for discovering effective information and potential models from a large amount of data.As an important part of data mining,cluster analysis is widely used in business,medicine,scientific research and other fields by grouping massive data to discover the characteristics However,personal privacy information is inevitably exposed in the process of cluster analysis.In order to solve privacy leakage problem,privacy protection technology came into being.Differential privacy,as a kind of privacy protection technology based on data scrambling,does not require the background knowledge of attackers,has become a research focus of privacy protection.At present,the availability of clustering algorithms is usually reduced due to differential privacy noise and sensitive input parameters in cluster algorithms based on differential privacy.Therefore,how to improve the clustering algorithm based on differential privacy and enhance the availability of data is the key to solve the problemIn the paper,a CFSFDP algorithm based on differential privacy is proposed to solve the problem that traditional cluster algorithm based on differential privacy is not suitable for nonconvex data sets and is greatly affected by input parameters.According to the characteristics of CFSFDP algorithm,differential privacy noise is introduced into the steps of density calculation and distance calculation respectively to make the algorithm meet the differential privacy model,and theoretical analysis and experimental comparison are performed to verify the security and availability.Secondly,DP-rcCFSFDP algorithm is proposed due to the poor performance of CFSFDP on uniform distributed data and the low clustering accuracy of the CFSFDP algorithm based on differential privacy.The selection of reachable center points is optimized by clustering the lower-density center points with the reachable and higher-density center points to improve the availability of clustering and reduce the impact of noise,which is verified by experiments.Finally,the DP-rcCFSFDP algorithm is applied to collaborative filtering recommendation system,which clusters in the data set,narrows the scope of nearest neighbor search,calculates similarity,generates prediction scores and recommends,to achieve a balance between privacy protection and recommendation accuracy.