基于共享逆近邻与指数核的密度峰聚类算法

Density peak clustering algorithm based on shared reverse nearest neighbors and exponential kernels

针对密度峰聚类算法中局部密度定义和聚类分配策略的不足,提出了一种基于共享逆近邻与指数核的密度峰聚类算法。该算法利用样本的共享逆近邻点和指数核函数构造一种相似度,得到一种新的密度并将其应用在密度峰聚类算法中生成初始簇,然后将这些簇与凝聚层次聚类算法结合形成最终的类簇。数值实验证明:提出的基于共享逆近邻与指数核的密度峰聚类算法在真实数据集上的聚类结果要优于其他密度聚类算法,并能有效解决密度峰聚类算法中局部密度定义问题和聚类分配策略问题。

For the deficiency of local density definition and cluster allocation strategy in the density peak clustering algorithm,a density peak clustering algorithm based on shared reverse nearest neighbors and exponential kernels is proposed.The proposed algorithm defines the similarity based on shared reverse nearest neighbors of the sample combined with the exponential kernel function,and then a new density is formed and applied in the density clustering algorithm to generate the initial clusters,which are combined with the agglomeration hierarchical clustering algorithm to form the final cluster.The numerical experiments show that the clustering results of the proposed algorithm on real datasets are better than other density clustering algorithms,and it effectively solves the problem of local density definition and cluster allocation strategy in density peak clustering algorithm.