Research On Privacy Preserving Association Rule Mining Based On Fully Homomorphic Encryption

With the rapid development of Information Technology, the global economy presents a trend of integration, the data is no longer stored in only one site, the majority of database is distributed, this means that data is distributed in two or more sites. For finding out the global association rules,that data owners want to have cooperation in data mining, but due to the problem of privacy, the participants do not want their private data to be known by others. Therefore, it is necessary to research an efficient algorithm of privacy preserving association rule mining with distributed data, which has important significance in theory and application.Association rule mining is an important research in data mining, and in this paper, we only discuss privacy preserving in association rule mining. Based on the comprehensive analysis of existing algorithm, this paper applies fully homomorphic encryption algorithm which meets both homomorphic add and homomorphic multiply, and optimizes the size of key and efficiency of encryption, combines it with multi-party computation, an algorithm of privacy preserving association rule mining based on fully homomorphic encryption is proposed, this algorithm encrypts data which transmissed between sites, and designs protocols with low cost of computation and communication for the operation of ciphertext, the global frequent itemsets can be determined by these protocols, and solves the issue of privacy preserving association rule mining finally. The main content of this paper are as follows:(1) This paper proposed an enhanced algorithm EDGHV based on integers. This algorithm design a somewhat homomorphic encryption first, with the help of compressed public key and decryption circuit, it can achieve bootstrapping. The new algorithm compares with DGHV has good performances and compress the size of public key effectively.(2) A horizontally distributed data oriented privacy preserving association rule mining based on EDGHV algorithm(EDGHV-HPP) is proposed. This algorithm generates all gloabl frequent itemsets by comparing the ciphertext of support of candidate itemsets. First EDGHV algorithm generates local frequent itemsets according to Apriori algorithm. To avoid leakage of privacy, algorithm encrypts the support of local itemsets ensures the security of data transmission between sites, comparing with this two numbers with protocols which designed for operation of homomorphic add and homomorphic multiply. Experiments show that compares with algorithm which based on Paillier, this new algorithm EDGHV-HPP can ensure the security and accuracy in association rule.(3) A vertically distributed data oriented privacy preserving association rule mining based on EDGHV algorithm(EDGHV-VPP) is proposed. This algorithm design a secure multi-party computation protocol which based on EDGHV, by this protocol of scalar product, it computes the support counts of candidates at each sites, then determines the itemset is frequent or not. Experiments show that this algorithm EDGHV-VPP achieves good performance, meanwhile, with low cost of communication and computation.(4) Privacy preserving association rule mining based on fully homomorphic encryption system is developed, and introduces the main classes of this system,finally, we test this system and show the result.