An Algorithm of Density Peak Clustering with Customized Cluster Quantity Based on Particle Swarm Optimization

Density peak clustering (DPC) is a widely used density-based clustering method. However, the clustering results obtained by DPC heavily depend on manually setting the cut-off distance, and DPC cannot generate a set of clusters with customized cluster quantity. An algorithm of density peak clustering with customized cluster quantity (CCQDPC) based on particle swarm optimization (PSO) is proposed. First, an evaluation factor is defined. The factor and customized cluster quantity are jointly regarded as the objective function of PSO, and the cut-off distance is the parameter to be optimized. Then, a set of constraints and optimization rules have been established. The two core constraints are as follows. The closer the number of clusters is to the customized quantity, the higher the quality of the particle. The larger the evaluation factor value, the higher the particle quality for particles with the same number of clusters. Finally, by iterating and evolving particles, the optimal cut-off distance is obtained, and the clustering results with customized cluster quantity and high evaluation are generated simultaneously. Seven common data sets are used for experiments. The findings demonstrate that the efficiency of CCQDPC is good, as it only requires a small particle swarm size and fewer iterations to obtain the expected clustering results. Compared to classical partitioning-based clustering algorithms, the CCQDPC algorithm has better clustering quality.