Research Of Private Preserving Data Mining Algorithm

Data mining has been used in many fields, such as financial, medicine and so on. Now more and more information can be acquired from the internet, and more and more tools of data mining have been developed, and so the data mining has been a threatened the security of the data and the privacy on certain extend. In order to preserve the privacy, more and more efficient privacy-preserving data mining algorithm are needed. Several private-preserving classified data mining algorithms are proposed in this paper. The main contents of the paper are described as follows.The research background of and the development status are introduced in the first chapter, and then the main algorithms of data mining are introduced in the second chapter. The main private-preserving data mining algorithm include disclosure control of the statistical methods, random methods, encryption technology, and so on.The private-preserving classified data mining algorithm based on Bayesian network is proposed in the third chapter. The algorithm is based on the partitioned database. And the algorithm is composed of two parts. One part is the local Bayesian network learning, and the other is global learning. The quantum genetic algorithm is used in the local Bayesian network structure learning to get the local Bayesian network structure. In order to preserve the local data, the secure sum algorithm is used on the local Bayesian network before it is sent to the honesty third party. The algorithm is proved to be feasible on both theory and experiment.In the fourth chapter, The Bayesian network learning is based on the quantum ant colony optimization algorithm. After the local Bayesian network learning, the secure sum algorithm is also used to protect the privacy as it is used in the third chapter. In the algorithm, the mutual information is used in the global Bayesian network. The algorithm is proved to be feasible both in theory and experiment.In the fifth chapter, the Bayesian network structure learning is based on modified swarm optimization algorithm. The data is discrete, and so the binary coding is used in the algorithm. The position of particles is binary, but the speed of the particle is continuous which represents the probability of the particle. The optimal particle is saved in the process of the algorithm, and the mutation operation is used to reduce the probability of getting the local optimal solution. In this private-preserving algorithm, the Rijndael encryption algorithm is used to protect privacy. During the global Bayesian network learning, the mutual information is used. The algorithm is proved to be feasible.In the sixth chapter, a modified private preserving SVM (MPPSVM) classification algorithm on distributed database is proposed. In which, the SVM is used for classification on distributed database, and the Gram matrix is used for creating the global model of SVM. To preserve data privacy, a modified secure sum is proposed in this chapter. This algorithm can protect the data from disclosing to other parties of the distributed database. With the increasing number of the collusion, the disclosure possibility may be increased, but the security can still be guaranteed by changing the coefficient. We did some experiments on the algorithm, and the results show that the algorithm protects the data privacy better than other algorithms.In the end, we summarize the main idea and the shortcomings of the dissertation, and supposed the advanced research direction.