支持差分隐私保护及离群点消除的并行K-means算法.

Aiming at the problem of privacy protection of clustering analysis in big data environment, based on the MapReduce computing framework, this paper proposed a parallel K-means algorithm that supported differential privacy protection and outlier elimination. The algorithm parallelly calculated the Euclidean distance matrix and nearest neighbor hypersphere radius between points in data set to derive the decision threshold of outliers, and then completed the initial cluster center selection and parallel clustering process under differential privacy protection. The theoretical analysis proves that the proposed algorithm satisfies ε-differential privacy, and the experimental results show that, compared with other algorithms, this algorithm performs better and has a good balance between the validity of privacy protection, the availability of clustering result and the efficiency of implementation. [ABSTRACT FROM AUTHOR]