Research On Privacy Preserving Algorithms For Association Rules Mining In Distributed Environment

With the data mining applied more and more widely in e-commerce,medical,etc,people pay more and more attention to the issues of privacy disclosure generated by data mining. How to cooperate for privacy preserving data mining has become a hot research in distributed environment. So far, distributed privacy preserving data mining algorithms can protect the data in a way, however, for the cost of higher computational complexity. So, how to reach a better balance between privacy and computational efficiency is still the goal.We design the privacy preserving association rule mining algorithm in horizontal distributed and vertically distributed environment,under the research and analysis of privacy preserving association rule mining algorithm at home and abroad.In horizontal distributed environment, we propose a distributed privacy preserving association rules mining algorithm (ARPRD), which based on multi-parameter random disturbance and homomorphism encryption, for the disadvantage of high computational cost of methods based on the encryption in currently. The idea of our algorithm is every distributed site uses multi-parameter random disturbed technique which is high efficiency and privacy to disturb the original data firstly, then uses paillier’s homomorphism encrypted method which complexity is constant to encrypt local support, and then data center uses property of paillier’s homomorphism encrypted to find out global support accurately,at last data center gets global frequent itemsets according to the value of minimum support, generates the association rules thereby. We analyze the ARPRD algorithm is safe and accurate in theory, experiments prove that ARPRD algorithm can improve accuracy of mining and computing efficiency while protecting privacy.In the vertical distributed environment, the key is how to calculate the global support of itemsets safely. The protocol to calculate the global support safely only apply to two parties involved and needs to generate a lot of random numbers and complex calculations, computational efficiency currently. So, we design a protocol called SMGSP that calculating the global support of itemsets safely and can apply to many parties. SMGSP based on the paillier’s homomorphism encryption technology to calculate the sum easily.And then proposed a vertical distributed environment privacy preserving association rules mining algorithm (PPVDR) based on the protocol of SMGSP. We analyze the PPVDR algorithm is safe and true and has good performance in theory,experiment results prove that the PPVDR reduced the cost of computation while the results of mining is accurate in the end.