SecMTSC: Achieving Privacy-Preserving and Efficient Multivariate Time Series Classification for Healthcare

Abstract

Time series data are widely used in many application scenarios. For example, in healthcare scenario, time series-based health predictions require users to submit their time series data, so that service provider (e.g., medical center) can predict the user's health status through calculations (e.g., classification). To relieve the pressure on local computing, service provider may outsource its data to the clouds. However, since clouds are not fully trusted, how to outsource data to the cloud without sacrificing data security but allowing service provider to remotely predict user's health status is challenging. To address this, we present Secure Multivariate Time Series Classification (SecMTSC), a privacy-preserving and efficient multivariate time series classification scheme, which can be used for healthcare and other applications. Specifically, by leveraging two-party secure computation protocols, the proposed scheme enables service provider to respond the classification query launched by query users on their private sequence without disclosing the data privacy. To further enhance query efficiency, we design a Secure Filtering Protocol, which can filter and extract valid data from raw data before formal calculation, thereby selecting relatively qualified candidate sequences. Security proofs have shown that our scheme is secure under the semi-honest model. Furthermore, experimental results have demonstrated the promising practicality of our scheme.