Efficient Privacy-Preserving Data Mining in Malicious Model

Abstract

In many distributed data mining settings, disclosure of the original data sets is not acceptable due to privacy concerns. To address such concerns, privacy-preserving data mining has been an active research area in recent years. While confidentiality is a key issue, scalability is also an important aspect to assess the performance of a privacy-preserving data mining algorithms for practical applications. With this in mind, Kantarcioglu et al. proposed secure dot product and secure set-intersection protocols for privacy-preserving data mining in malicious adversarial model using zero knowledge proofs, since the assumption of semi-honest adversary is unrealistic in some settings. Both the computation and communication complexities are linear with the number of data items in the protocols proposed by Kantarcioglu et al. In this paper, we build efficient and secure dot product and set-intersection protocols in malicious model. In our work, the complexity of computation and communication for proof of knowledge is always constant (independent of the number of data items), while the complexity of computation and communication for the encrypted messages remains the same as in Kantarcioglu et al.’s work (linear with the number of data items). Furthermore, we provide the security model in Universal Composability framework.