DPLK-Means: A Novel Differential Privacy K-Means Mechanism

K-means algorithm is an important type of clustering algorithm and the foundation of some data mining methods. But it has the risk of privacy disclosure in the process of clustering. In order to solve this problem, Blum et al. proposed a differential privacy K-means algorithm, which can prevent privacy disclosure effectively. However, the availability of clustering results is reduced due to the added noise. In this paper, we propose a novel DPLK-means algorithm based on differential privacy, which improves the selection of the initial center points through performing the differential privacy K-means algorithm to each subset divided by the original dataset. Performance evaluation shows that our algorithm improves the availability of clustering results compared to the existing differential privacy K-means algorithm at the same privacy level.