Privacy Preserving Support Vectormachine Based On Orthogonal Transformation And Secure Dot-Products

Privacy preserving data mining has become a hot issue in academic circles since the 90s in the last century. The disclosure of data information appears in data mining tasks is studied by a number of scholars, making privacy preserving data mining, one aspect of which is privacy preserving support vector machine, come into being. The achievements of privacy data mining and support vector machine are briefly introduced in this paper. And new privacy preserving methods are proposed to solve the problem of privacy disclosure when support vector machine model is established. The main results of this paper is as follows.Data privacy will be disclosed as the constructing of support vector machine model needs direct access to data. By using the property of orthogonal transformation a new privacy preserving support vector machine model is proposed. Moreover, algorithms corresponding to three forms of data partition are put forward. It is pointed out that the algorithm on checkerboard partitioned data can be applied to an arbitrarily partitioned data as it can be transformed to a checkerboard one.The computation of dot-product of vectors is key in constructing the dual model of standard support vector machine. Dot-product of data vectors held by two parties can be obtained by applying secure dot-product protocols under the premise that data information is hidden.The security and exactness of secure dot-product protocol by Ioannidis is discussed in the paper. And a privacy preserving support vector machine algorithm is proposed, applying the protocol to solve the dual model of support vector machine.