Research On Privacy-Preserving Data Mining Of Cluster In Cloud Environment

With the explosion of data,it is more difficult for users who have short resources to perform data mining locally.The users choose to outsource the data mining tasks to cloud services with the strong power of computing and storage space to increase the efficiency and save cost.However,some sensitive information of data mined is revealed in cloud environment.Therefore,how to mine data without destroying data privacy is extremely important,which has become a research hotspot.This paper will combine the knowledges of cryptography to study the privacy-preserving clustering data mining,and mainly doing the following two aspects:1.A dynamically adjustable privacy-preserving clustering data mining scheme is proposed.The scheme allows the cloud server to dynamically adjust the number of clusters and appropriately select the initial clustering centers on the encrypted datasets,which improves the clustering efficiency while ensuring the data privacy and the clustering accuracy.In privacy-preserving clustering schemes,since the k-means clustering needs to determine the number of clusters and the initial clustering centers in advance,it requires a large amount of computation for inexperienced data owners.Therefore,we deliver these tasks to the cloud servers with a strong power of computation to reduce the computing burden of data owners.The data owner uploads data that is encrypted by utilizing a lightweight symmetric encryption algorithm to the cloud server.The cloud server dynamically adjusts the number of clusters,and sends the last number of clusters and the initial clustering centers to data owners.The scheme reduces the computation of data owners while the clustering accuracy has not changed through the analysis and proofs,and it also proves that our scheme is safe and efficient.2.A more secure outsourced k-means clustering scheme under multiple keys is proposed.The scheme allows data owners to securely outsource data to two non-collusive cloud servers by using their own public and private keys and allows the cloud server to handle encrypted data under different keys.In order to ensure the data privacy,we use the additively homomorphic cryptosystem with the double decryption properties to encrypt data.Then,a secure multiplication protocol is constructed.The two non-collusive cloud servers participating in protocol can perform multiplication over the ciphertexts without revealing any information about the original data.We construct a set of sub-protocols by utilizing the additive and multiplier protocols to complete the privacy-preserving clustering data mining,for example,the secure ciphertexts transformation protocol,secure squared Euclidean distance protocol and other sub-protocols.In addition,the data owner uses a symmetric cryptosystem to re-encrypt the encrypted database to the cloud server,preventing the cloud server with the master key from decrypting the ciphertexts of data owners.The data owners and the querying owners do not participate in mining process.Our scheme is proved to be secure by analyzing and discussing sub-protocols.Finally,we compare the current typical privacy-preserving clustering data mining schemes,which shows that our scheme is optimal.