Density Peak Clustering Using Grey Wolf Optimization Approach

Density peak clustering (DPC) finds the center of the cluster as the point with high density and a large distance from the center of the other clusters. However, DPC requires manual intervention to choose the centers of the cluster in the decision graph and is highly dependent on the user-defined parameter $$\mu _{c}$$ μ c , which decides the cut-off value to find the density around the given point. This study provides a clustering algorithm based on the grey wolf optimization strategy which aims to fix the cluster center of the density peak clustering without human involvement. Firstly, a fitness function is constructed, inversely proportional to the density and distance of the points from the centers of the cluster. Secondly, the cluster centers representing the peaks are randomly initialized and minimized by the grey wolf optimizer using the defined fitness function which calculates the density of each point and the distance among them. Third, the user-defined parameter $$\mu _{c}$$ μ c which is highly dependent on the data size in the DPC is automatically determined using the Gaussian distribution. Finally, to check the stability of the optimal cluster centers, the centroid method is applied and the number of iterations is recorded to observe the change in cluster centers. To investigate the performance of the proposed approach, simulations on real-world and synthetic data are performed and evaluated using different clustering evaluation indices. The finding shows that the cluster centers obtained are more stable than those found by k-means, k-medoids, density-based spatial clustering (DBSCAN), k-nearest-density peak clustering (KNN-DPC), shared-nearest-neighbor-density-based clustering (SNN-DPC), particle swarm optimization clustering (PSO-CC), and grey wolf optimization clustering (GWO-CC).