PRIVACY-PRESERVING OUTLIER DETECTION OVER ARBITRARILY PARTITIONED DATA

Outlier detection has a wide range of applications, such as electronic commerce, credit card fraud detection and terrorism prediction which is related to the national security. Due to the importance of security and privacy, the study of distributed data mining is even more useful. Previous privacy-preserving outlier detection protocols over both horizontally and vertically partitioned data fail to deal with the case when the data is arbitrarily distributed among multi-party, and also leak information about the participants' privacy data. In this paper, we present a privacy-preserving distance-based outlier detection protocol over arbitrarily partitioned data without any information leakage. The protocol can also be applied to k-nearest neighbors mining. The security of our protocol is proved based on Secure Multi-party Computation theory.