基于密度权重的优化差分隐私K-medoids聚类算法

Optimal differential privacy K-medoids clustering algorithm based on density weights

K-medoids算法作为数据挖掘中重要的一种聚类算法,与差分隐私保护的结合有助于信息数据的安全,原有的基于差分隐私保护的K-medoids聚类算法在初始中心点的选择上仍然具有盲目性和随机性,在一定程度上降低了聚类效果。本文针对这一问题提出一种基于密度权重的优化差分隐私K-medoids(DWDPK-medoids)聚类算法,通过引入数据密度权重知识,确定算法的初始中心点和聚类数,以提高聚类效果和稳定性。安全性分析表明,算法满足ε-差分隐私保护;通过对UCI真实数据集的仿真实验表明,相同隐私预算下该算法比DPK-medoids具有更好的聚类效果和稳定性。

As an important kind of clustering algorithm in data mining,the combination of K-medoids algorithm and differential privacy protection helps the security of information data.However,the original K-medoids clustering algorithm based on differential privacy protection is still blind and random in the selection of initial centroids,which reduces the clustering effect to some extent.To address this problem,an optimal differential privacy K-medoids(DWDPK-medoids)clustering algorithm based on density weights is proposed to determine the initial centroids and the number of clusters of the algorithm by introducing the knowledge of data density weights to improve the clustering effect and stability.The security analysis shows that the algorithm satisfies-differential privacy protection;the simulation experimental results on real UCI datasets show that the algorithm has better clustering effect and stability than DPK-medoids under the same privacy budget.