Differential privacy k-means clustering method based on cluster similarity and transformation invariance

Abstract

The invention discloses a differential privacy k-means clustering method based on cluster similarity and transformation invariance. The differential privacy k-means clustering method comprises the following steps: preprocessing a data set; reversely sorting the probabilities P that the sample is selected as the next initial cluster center, and selecting K initial cluster centers by using a roulette method; calculating the product of the Euclidean distance from each sample to K cluster centers and the similarity to each cluster set, and taking the product as a similarity measurement index; dividing clusters according to similarity measurement indexes; calculating a new cluster center u'j, and performing privacy protection on the new cluster center by using a differential privacy Laplace noise mechanism; correcting the new cluster center disturbed by the Laplace noise mechanism according to the transformation invariance of the differential privacy; and repeating the steps until a convergence condition is met or the maximum number N of iterations is reached. According to the method, the problem that the result availability of an existing differential privacy clustering method is poorafter Laplace noise disturbance is added is solved.

Description

technical field [0001] The invention belongs to the technical field of privacy protection, and in particular relates to a differential privacy k-means clustering method based on cluster similarity and transformation invariance. Background technique [0002] As a typical unsupervised data mining method, K-means clustering can mine unknown knowledge and potential value from massive data. However, while mining useful information, personal privacy information in the data may be leaked, causing immeasurable threats and losses to users. Therefore, in the process of k-means clustering analysis, it is necessary to protect the personal privacy information in the data and ensure the availability of the final clustering results. [0003] As attackers have more and more background knowledge, traditional privacy protection technologies, such as k-anonymity, t-closeness, etc., have become difficult to guarantee data security. As a new privacy protection model with strict mathematical pr...

AI Technical Summary

Problems solved by technology

[0004] The embodiment of the present invention provides a differentially private k-means clustering method based on cluster similarity and transformation invariance, aiming to solve the problem of poor availability of results after adding Laplace noise to existing differentially private clustering methods

Images