Research On Key Technologies Of Secure Data Mining Outsourcing In Cloud Computing Environment

With the rapid development of cloud computing technology,cloud service providers can offer powerful and flexible data storage and computing resources to users.Outsourcing data mining tasks to cloud can greatly reduce the operation and maintenance costs of users.However,due to various security threats such as data leakage in the cloud environment,and cloud service providers are not completely trusted by users,there are risks of privacy leakage of the outsourced data.Therefore,guaranteeing the security of data and results during the process of outsourced data mining in cloud environment has become a key problem in the field of cloud computing security,which is required to be solved urgently.The existing secure data mining outsourcing schemes usually cannot satisfy the requirements of high security and efficiency simultaneously,and might require the online participation of users during the outsourced computation processes,which affects the applicability of the schemes.To solve these problems,based on the comprehensive analysis of secure data mining outsourcing problem in cloud computing environment,we focus on the research of secure outsourcing technologies of k-nearest neighbor classification,deep neural network classification,k-means clustering and frequent itemset query.We establish reasonable system models and security models,and design secure and efficient outsourcing schemes without online users,which improves the practicability of the schemes.The main work and innovations of this paper are given in the following aspects:1.Considering the deficiencies on security,efficiency and applicability of the existing secure outsourcing schemes of k-nearest neighbor classification,we propose a secure outsourcing scheme of k-nearest neighbor classification using an efficient encryption algorithm,which supports efficient k-nearest neighbor classification over large encrypted databases.The main innovations of this scheme include:(1)We design the protocols of generation of database encryption key,database encryption,generation of query encryption key,query encryption,generation of re-encryption key for classification labels,k NN classification over encrypted database and classification labels decryption,and propose the secure outsourcing scheme of k-nearest neighbor classification based on these protocols.(2)We provide detailed security analysis of the proposed scheme under the semi-honest security model,proving that the scheme can protect database security,confidentiality of decryption key and query privacy,as well as hide data access patterns.(3)We analyze the performance through extensive experiments,and the experiment results show that the proposed scheme is very efficient.2.Considering the deficiencies that existing secure outsourcing schemes of k-nearest neighbor classification cannot simultaneously satisfy semantically secure ciphertexts and low computation cost,we propose a secure outsourcing scheme of k-nearest neighbor classification using hybrid public-key encryptions,which supports efficient k-nearest neighbor classification over semantically secure hybrid encrypted database.The main innovations of this scheme include:(1)We propose a secure inner product(SIP)protocol to compute the encrypted distances,which needs less computation complexity compared with the existing protocol for encrypted distance computation.(2)We propose a collusion resistant re-encryption key generation protocol,which protects the decryption key from cloud servers even if they collude with query users.(3)We propose an encrypted k NN classification protocol by using SIP and proxy re-encryption protocols as sub-routines,which securely computes the encrypted k NN classification labels in an efficient way.(4)We provide detailed security proofs and performance evaluations with extensive experiments.The experiment results show that the proposed scheme is very efficient.3.Considering the deficiencies on ciphertext classification accuracy and efficiency of the existing secure outsourcing schemes of deep neural network classification,we propose a secure outsourcing scheme of deep neural network classification using fully homomorphic encryption,which satisfies the requirements of high security,efficiency and classification accuracy.The main innovations of this scheme include:(1)We propose parametric polynomial activations,which adaptively learn the parameters of polynomials during the training process,and achieve better classification performance.(2)We design three small convolutional neural networks,which guarantee the efficiency of classification over ciphertexts.(3)We use fully homomorphic encryption to encrypt the data of users,which ensures data security.The processes of outsourced classification are performed by cloud server without online participation of users.(4)We provide performance analysis with extensive experiments.The experiment results show that the proposed scheme satisfies the requirements of high accuracy and efficiency simultaneously.4.Considering the deficiencies that existing secure outsourcing schemes of k-means clustering cannot simultaneously support semantically secure ciphertexts and high efficiency,we propose a secure outsourcing scheme of k-means clustering using fully homomorphic encryption,which supports efficient clustering over semantically secure encrypted database.The main innovations of this scheme include:(1)We design protocols of database encryption,secure computation of scaling factors,secure scaled square Euclidean distance,secure cluster update,secure computation of termination condition and secure cluster centers decryption,and propose the secure outsourcing scheme of k-means clustering based on these protocols.(2)We utilize the ciphertext packing technique to achieve efficient parallel computations during the clustering process of ciphertexts,which significantly improves the efficiency of the scheme.The whole outsourced clustering processes are performed by cloud servers without online participation of users.(3)We provide detailed security analysis of the proposed scheme under the semi-honest security model,proving that the scheme can protect database security and privacy of clustering results,as well as hide data access patterns.(4)We analyze the performance through extensive experiments,and the experiment results show that the proposed scheme can efficiently execute k-means clustering over large encrypted database.5.Considering the deficiencies on security,efficiency and applicability of the existing secure outsourcing schemes of frequent itemset mining,we propose a secure outsourcing scheme of frequent itemset query using hybrid homomorphic encryptions,which supports efficient frequent itemset query over semantically secure encrypted database.The main innovations of this scheme include:(1)We design the protocols of transaction database encryption,itemset encryption,secure frequent itemset mining and secure support decryption,and propose the secure outsourcing scheme of frequent itemset query based on these protocols.(2)We utilize the ciphertext packing technique to achieve efficient parallel computations during the mining process of ciphertexts,which significantly improves the efficiency of the scheme.The whole outsourced mining processes are performed by cloud servers without online participation of users.(3)We provide detailed security analysis of the proposed scheme under the semi-honest security model,proving that the scheme can protect database security and privacy of query results,as well as resist frequency analysis attacks.(4)We analyze the performance through extensive experiments,and the experiment results show that the proposed scheme can efficiently execute frequent itemset query over large encrypted database.