A Density Peak Clustering Method based on Improved Siphon Effect

Abstract

To solve the problem that the density peak clustering algorithm needs to manually select the clustering center, a density peak clustering algorithm based on improved siphon effect (IDPC) is designed. IDPC completes the clustering of research objects with various distribution shapes by removing the maximum cluster center weight in the descending sequence, calculating and iterating the potential difference, and selecting the relative equilibrium point. A comprehensive comparative experiment is carried out on UCI data set. The experimental results show that IDPC can accurately and automatically select the cluster center and cluster number, improve the performance indexes such as Fowles mallows score, Rand index and adjusted mutual information, and can adaptively process data sets with low dimension and various distribution shapes.