Privacy Preserving Association Rule Mining Scheme Based On Homomorphic Encryption

With the rapid development of information technology,both the update rate of data and the types of new data are increasing,which makes our analysis and research of data more complicated,and it also marks the arrival of “big data era ”.With the advent of the era of big data,data mining has been widely used in various fields and has become a hotspot in the development of science and technology.Under such a background,it is difficult for resource-limited users to perform intensive data mining tasks locally.Therefore,it is a reasonable choice for users to deliver data mining tasks to cloud servers with large storage space and strong computing capabilities.However,if proper privacy protection measures are not taken for the analysis of data,data mining will be the source of the nightmare of user privacy leaks.Therefore,data mining under privacy protection comes into being,data mining is carried out under semi-honest model,and entities may be curious about users' privacy information.Moreover,due to the transparency of the data transmission channel,data may also be eavesdropped during transmission,and the legality of the data user's identity is also deceptive.Therefore,how to design a privacy-preserving association rule mining scheme with both security and reliability has become a challenge that must be faced directly.This thesis mainly focuses on the above issues as follows:Our first work is to conduct a security attack on “Privacy-preserving association rule mining for horizontally partitioned healthcare data: a case study on the heart diseases”proposed by Nikunj et al.,and propose a safer disease diagnosis scheme.Concretely,this thesis proves that the security of Nikunj et al.'s scheme can not reach the strength they claim through three aspects: irrationality of parameter generation in initialization stage,the insecurity of ciphertexts transmission and the possibility of mining participants' collusion.In this way,we remind future scholars to avoid such problems in the process of designing schemes and carry out better scientific research.In addition,in the context of this research,we propose an improved solution 1,namely the privacy preserving disease analysis solution based on the electronic medical system.We adopt the homomorphic cryptographic scheme Paillier with semantic security to realize privacy preserving data mining,and the private key of Paillier is weakened and then distributed to each mining entity.This solution can resist the collusion of data users,and also perfectly avoids the privacy leakage problem of the original solution.Finally,by analyzing the safety and correctness of the improved solution 1 and performing performance tests,it is shown that the improved scheme 1 is safe and feasible.Our second work is to propose a privacy-preserving commodity relevance mining scheme based on transaction data.Our scheme does not have the maximum confidential authority which overcomes the shortcoming that the cloud server Evaluator has the strong private key in the prior art that cannot resist the active interception attack of the Evaluator during the data outsourcing phase.At the same time,our solution adopts a batch-verifiable short digital signature scheme BLS to ensure the integrity of the transmitted data and the identity of the sender,which overcomes the shortcomings of the unknown data source in the prior art,and makes the scheme avoid the malicious attacker's poison attack.Compared with the existing scheme,our solution has the advantages of higher security and more reliable mining results.