A Grid-Based Density Peaks Clustering Algorithm

This article focuses on the improvement of density peaks clustering (DPC, also known as clustering by fast search and find of density peaks) by the introduction of the grid clustering. A grid division is used to divide the data space into grid cells. The global characteristic of grid cell replaces the characteristic of the data points. The local density of DPC is replaced by the density of the grid cell. The cut-off distance of DPC is no longer needed. Then, the cluster centers are determined adaptively based on the quantities. Finally, a two-stage allocation strategy (neighborhood searching and border grid cell assigning) is introduced to obtain the final clustering results. Six datasets are used to experimentally verify the effectiveness and performance of the proposed algorithm. Experimental results show that the improvement of DPC is successful. Compared with DPC, the proposed algorithm is more efficient with less manual intervention.

I. Introduction

As a key technique of pattern recognition, clustering analysis plays an important and broad role and has been widely applied in many fields [1], [2], [3], including image processing/machine vision, robot sensing, statistics, bioinformatics, data mining, and machine learning. Many clustering methods have been proposed and great technical progresses have been achieved [1], [2], [3], [4], such as K-means, density-based spatial clustering of applications with noise (DBSCAN), spectral clustering, and so on. However, with the rapid development of modern technology, the requirements of pattern recognition become higher and higher. The conventional clustering methods could not meet the increasing requirements in modern society.