基于t-SNE降维的密度峰值聚类算法

Density Peak Clustering Algorithm Based on t-SNE Dimensionality Reduction

为了提高密度峰值聚类(DPC)算法处理复杂高维数据的能力,提出了一种基于t-SNE降维的密度峰值聚类算法(t-SNE-DPC)。该算法用t-SNE算法对数据进行预处理,将高维数据点间的关系用概率分布映射到低维空间中,通过最小化相对熵最大化保留数据的本质特征,使用密度峰值聚类算法进行聚类操作。仿真实验结果表明,t-SNE-DPC可以高效地对高维数据进行聚类,在AMI指标上的聚类结果可达0.828。

In order to improve the ability of Density Peak Clustering(DPC)algorithm to deal with complex high-dimensional data,a density peak clustering algorithm is proposed based on t-SNE dimensionality reduction(t-SNE-DPC).The algorithm uses the t-SNE algorithm to pre-process the data,maps the relationship between high-dimensional data points to the low-dimensional space with probability distribution,maximizes the retention of the essential characteristics of the data by minimizing the relative entropy,and finally uses the density peak clustering.The algorithm performs clustering operations.The simulation results show that t-SNE-DPC can efficiently cluster high-dimensional data,and the clustering results of t-SNE-DPC on the AMI index can be as high as 0.828.