Privacy-preserving SVM of horizontally partitioned data for linear classification

When we use support vector machine (SVM) to solve the classical classification problem, we should know all data. However, the data sometimes can reveal private information which is protected by laws. So recently, there has been growing focus on finding solutions to get a SVM classifier without revealing any information of the privately-held data. In this paper, we propose a new method which is ameliorated from the usual SVM to solve this problem over horizontally partitioned data which can protect the private information of the data completely. And under some special conditions, the model provided in this paper can achieve same accuracy with the usual SVM constituted by the original data. The experiments on real datasets show that the classification accuracy of our proposed method on the protected data is approximate to the SVM classifier on the original data.