Privacy Preserving Outlier Detection Algorithm Based On Secure Multi-party Computation

Outlier detection and privacy preserving are important research directions in the field of data analysis and data mining.With the development of information technology,data sources are more diversified.When data is distributed at different sites,based on legal and privacy considerations,it is necessary to design a corresponding privacy preserving scheme to jointly detect outliers at each site while protecting the privacy of each site data.Based on the above requirements,this paper proposes privacy protection outlier detection algorithms based on secure multi-party computation.The main research work includes the following two aspects:1.For the vertically distributed data source,this paper proposes a domain-based privacy protection anomaly detection algorithm.The algorithm aims to efficiently detect outliers while protecting each participant's private data sets.According to the domain-connected outlier detection algorithm ODCD,calculating the connected radius based on the global to connect the data points,judged outlier points which are not connected after scanning the whole dataset.And outlier clusters are judged which are connected by the radius.Then the algorithm is extended to distributed computing.The Paillier homomorphic encryption technology and the data perturbation matrix are used to design a secure multi-party computing protocol,which ensures the security of data transmission and interaction,effectively resists multi-party collusion attacks,and simplifies encryption and decryption operations.According to the experimental comparison and analysis of security protocols,it is verified that the proposed algorithm can efficiently detect outliers and guarantee the private data security of each participant.2.This paper proposes AVF-based privacy preserving outlier detection algorithms.According to the characteristics of horizontal distribution,the AVF-based outlier detection algorithm under horizontal distribution is proposed.Using the candidate outlier judged conditions,the obtained candidate outliers are equally eliminated,and the final outliers are obtained.The BP protocol is extended to a secure multi-party computing to design the AVF-based privacy preserving outlier detection algorithm.According to the vertical distribution of data,secure multi-party calculations under vertical distribution are designed using security sum protocols.The accuracy and effectiveness of the proposed outlier detection algorithm under horizontal distribution and vertical distribution are verified by comparison with centralized data outlier detection.The effectiveness of the proposed algorithm is verified on simulation dataset and UCI test dataset,and compared with the classical outlier detection algorithm.The recall ratio and false ratio of performance indexes of the outlier detection are shown the advantages of the proposed algorithms.This paper uses secure multi-party protocols to design a privacy protection scheme,and analyzes the proposed privacy preserving outlier detection algorithms from accuracy,security and efficiency.