Ensemble Clustering Algorithm Based on Cluster Space Features Under Local Differential Privacy

Aiming at the shortcomings caused by the limitations of a single clustering method in the existing differential privacy protection clustering algorithm research, an ensemble clustering algorithm based on cluster spatial characteristics under local differential privacy was proposed. Firstly, a data coding perturbation method based on multi-tree was proposed to realize personalized perturbation through one-time interaction, reduce the communication cost of data transmission and reduce the amount of individual noise injection. Secondly, a data aggregation method based on the idea of ensemble is designed. Through a new similarity standard based on spatial characteristics and size between clusters, the super-clustering technology is used to realize multi-model ensemble clustering, which ensures the high quality of clustering when the proposed clustering algorithm deals with different types of data. Finally, the effectiveness of the algorithm is verified by experiments.