EPCS: an efficient and privacy‐preserving classification service query framework for SVM
Hui Zhu, Xiaoxia Liu, Rongxing Lu, Hui Li · Security and Communication Networks · 2016
Abstract Support vector machine (SVM) classification, which can handle large data sets in high dimensional spaces, has been widely applied in numerous settings nowadays, such as genetic match, spam detection, and financial prediction. However, because of the data's sensitivity and classifiers' confidentiality, how to provide a privacy‐preserving SVM classification has attracted considerable interest recently. Aiming at these privacy challenges, in this paper, we present an efficient and privacy‐preserving classification service query framework, named EPCS, for the linear kernel SVM classifier. Specifically, based on lightweight multiparty random masking and polynomial aggregation techniques, EPCS preserves the privacy of users' data and SVM classifier efficiently during the process of user query. Through detailed security analysis, we show that EPCS can resist various security threats. In addition, we implement EPCS in Java and evaluate EPCS over a smart phone and a PC, and extensive simulation results demonstrate that the EPCS framework can indeed achieve high efficiency and effectiveness in SVM classification. Copyright © 2016 John Wiley & Sons, Ltd.