Efficient and Privacy-Preserving Outsourced SVM Classification in Public Cloud

Data classification has become an important and prevailing technique for big data analytics. Typically, a data classifier is designed and outsourced to a public cloud. A service provider then can easily provide various services and handle frequent and massive classification requests from users. With privacy concerns as well as Intellectual Property(IP) protection issues, the valuable classifier and the sensitive user data cannot be directly exposed to the public cloud. In this paper, we focus on the Support Vector Machine (SVM), one of the most popular classifiers, and propose an efficient and privacy-preserving outsourcing scheme for SVM classification in public clouds. Specifically, the service provider is allowed to transform the traditional SVM classifier to fixed hyper-rectangles and the order-preserving encryption is utilized to encrypt these hyper-rectangles as the encrypted classifier. Afterwards, the encrypted classifier is outsourced to the public cloud, and a user can submit an encrypted range query to the cloud and obtain the classification results back. Security analysis and extensive experimental evaluation demonstrate that our scheme can protect the confidentiality of classifier and users' data and achieves efficient SVM classification in terms of computational cost.

I. Introduction

As one of important big data analytic techniques, data classification can be applied to make accurate predictions for unknown data samples, which benefits a large variety of applications, such as online disease diagnosis [1], [2], and vehicular intelligence applications [3]. In general, different kinds of data classification functions (a.k.a. classifiers), owned by a service provider, can be outsourced to a third-party cloud server for easy configuration and convenient management, meanwhile users’ service quality can also be improved due to the cloud’s powerful computation and storage resources. Support Vector Machine (SVM) is one of the most widely used classifiers, especially for online diagnosis services. For instance, a hospital can outsource its SVM diagnosis classifier to the public cloud to provide online diagnosis services, and patients can submit their biomedical information to the cloud and later obtain the corresponding diagnosis prediction.