| With the development of big data and cloud computing,the demand for cloud computing services is growing dramatically.When users apply for cloud computing services,their privacy data needs to be stored and computed on cloud platforms,which may cause leakage of private data.If traditional encryption schemes are used for privacy protection,the security of users’ privacy data calculation still cannot be guaranteed.Homomorphic encryption can not only encrypt data,but also directly operate ciphertexts.The new ciphertext obtained after decryption is the result of the operation.Therefore,homomorphic encryption can better protect users’ private data than traditional encryption schemes.According to the properties of homomorphic encryption scheme BGV that supports the operation among ciphertexts or between plaintexts and ciphertexts,we adopt a private data classification model based on C/S structure and designed four classic machine learning algorithms.They are hyperplane classifier,decision tree,KNN classifier,and K-MEANS classifier,which implement the secure two party classification calculation scheme based on semi honest application environment.The design points and calculation steps of these classifiers are described in detail,and their correctness and security are demonstrated rigorously.To solve the problem of low prediction accuracy caused by using encrypted samples to train the model,we train the model in plaintext and use the plaintext model to classify the encrypted samples,which improves the prediction accuracy more than 90%.To solve the problem of too many communication rounds and low efficiency caused by factors such as encryption protocol conversion,we only use BGV encryption scheme,combined with the SIMD technology that can batch calculate encrypted data and BGV ciphertext comparison protocol,the server can compute results without additional data provided by user.The two parties can complete classification in 1to 2 rounds of interaction and improves the efficiency of classification.The comparison experiment between our protocols and other protocols designed based on homomorphic encryption shows that our protocols have better prediction accuracy,operation efficiency and communication rounds,which can basically meet the needs of the given scenarios. |
