Research On Privacy-Preserving Classification Service Query Mechanism For SVM

As the era of big data comes,Support Vector Machine(SVM)gradually becomes a common data mining technique and has been widely applied in many real-world applications,such as image classification,handwriting recognition,genetic match,spam detection and financial prediction.SVM has attracted many scholars' attention both at home and abroad in recent years.However,with the bad effects of privacy,more and more people pay more attention to the protection of personal privacy.As for as privacy-preserving SVM classification service query schemes,how to provide efficient privacy-preserving classification query service without the leakage of users' data and SVM classifier is the research focus of many scholars.Many existing schemes based on homomorphic encryption techniques preserve the privacy of users' data and SVM classifier,but they are not very efficient and not appropriate for providing efficient privacy-preserving classification query service.Aiming at these problems,we study efficient privacy-preserving classification service query mechanism and propose efficient privacy-preserving linear and non-linear SVM classification service query schemes.Meanwhile,we implement these two schemes and deploy them in real environment to evaluate their integrated performance.Specifically,the emphases of this paper are as follows:First,to ensure users' data privacy and the confidentiality of SVM classifier,we improve the expression of linear SVM classifier and present an efficient and privacy-preserving linear SVM classification service query scheme based on multiparty random masking and polynomial aggregation techniques,which preserves the privacy of users' data and SVM classifier efficiently during the process of user query.Moreover,we analyze the computation complexity of this proposed scheme.Second,we also improve the expression of non-linear SVM classifier and construct an efficient and privacy-preserving non-linear SVM classification service query scheme based on lightweight polynomial aggregation technique,which preserves the privacy of users' data and SVM classifier efficiently during the process of user query as well.Besides,we analyze the computation complexity of this proposed scheme as well.Third,to evaluate the feasibility of our proposed schemes,we verify the correctness of both of these schemes.Moreover,through detailed security analysis,we show both of these algorithms can ensure that users' query information and service provider's prediction model are kept confidential,and has significantly less computation and communication overhead than existing schemes.