A Privacy Preserving Outlier Detection Algorithm Based on Connected Domain

Outlier detection and privacy preserving are the hot issues in data mining area. In this paper, a privacy preserving outlier detection algorithm based on connected domain is proposed in vertically distributed data models. The proposed algorithm aims at improving the recall ratio and reducing the fallout ratio of the outlier detection, while protecting the private data of each participant. After calculating the local distance matrix of all the pairs of objects in each party, a radius is selected to form a connected domain. In addition, the judgment of scattered outliers and outlier cluster are introduced. This algorithm is simple to understand and can be easily excuted, which needs no artificial parameters. The Paillier homomorphic encryption and data perturbation are used to design the secure multi-party computation protocol. This protocol can prevent information leakage, reduce communication complexity and simplify operations of encryption and decryption.