Research On Vertically Partitioned Data Oriented Privacy Preserving Data Mining Algorithm

With the rapid development of networking, communications technology, data mining takeson the features of large datasets and distributed mining. How to preserve data privacy andprevent sensitive information from disclosure has become a great challenge. Privacy PreservingData Mining (PPDM) has become an increasingly important issue in data mining datamining(DM) field.Clustering mining and association rule mining are widely used data mining methods. In thisthesis, the methods of privacy preserving for clustering mining and association rule mining invertically distributed environment are studied, and the purpose of the study is to mine thepotential knowledge in the database and protect data privacy at the same time.This thesis introduces several privacy preserving technologies which are commonly used indata mining from two aspects: the protection of sensitive data and the protection of sensitiveknowledge. On the basis of the overview of data mining technology, the thesis introducestheories of distributed clustering mining and association rule mining, analyzes several popularexisting distributed clustering mining and association rule mining algorithms and theiradvantages and disadvantages. Then, the thesis has done a lot of further research work forpreserving sensitive knowledge in vertically distributed clustering mining and association rulemining, which are as follows：(1) For the privacy preserving of clustering mining, we designed a new clustering algorithmVPPDK（Vertically Partitioned Data Oriented Privacy Preserving Distributed K-means）. Thealgorithm combines data interference and secure multiparty computation and uses secure averageprotocol to achieve privacy preserving clustering mining on vertically distributed data. It can notonly protect data privacy, but also effectively get mining results(2) For the privacy preserving of association rule mining, we propose a new CryptologyBased Strategy for Privacy Preserving Association Rule Mining，named CRYPPARM. InCRYPPARM, We use secure scalar product protocols and public key cryptosystems to efficientlymine association rules over vertically partitioned data. We also introduce a partial topology toreduce communication cost as much as possible.(3) In order to verify the algorithms designed by this thesis, experiments are done, and the experimental results show that those algorithms can achieve improvements in terms of privacy,accuracy, and efficiency.