A Cryptographic Solution to Privacy-Preserving Two-Party Sign Test Computation on Vertically Partitioned Data

Abstract

Statistical hypothesis test is an important data analysis theory that has found applications in various research fields. It provides a theoretical foundation to determine whether sufficient evidence exists to reject hypothesis for the trial using experimental results, in order to make a decision. In this paper, we address one of the fundamental test theories: the Nonparametric Sign Test, under the privacy-preserving context. In this context, two parties, Alice and Bob, would like to perform a sign test on their joint dataset, but neither of them is willing to disclose their private raw data to the other party. More specifically, this paper addresses the problem where the joint dataset consists of two vertically partitioned datasets. Our previous work [14] has addressed this problem using data perturbation techniques. However, in a case when the privacy of individual data objects and data subjects are of high concern, using data perturbation techniques may not be sufficient. This paper proposes an alternative solution to this problem by employing an additive homomorphic encryption scheme and an on-line STTP. We show that this solution can offer better privacy preservation, in terms of individual data confidentiality and individual privacy, than the previous solution. A security comparison with TTPV and P22NSTV solutions is also presented.