Density Peaks Clustering Algorithm Based on Weighted Shared Nearest Neighbors

Density Peaks Clustering is a commonly utilized, effective, and non-iterative algorithm for clustering. However, studies have demonstrated that DPC faces challenges when dealing with non-spherical clusters and datasets with uneven density distributions, particularly in selecting appropriate cluster centers and its sensitivity to the cutoff distance parameter. To overcome the limitations of DPC in handling uneven density distributions, this paper introduces a Density Peaks Clustering Algorithm Based on Weighted Shared Nearest Neighbors. This approach utilizes natural neighbors to determine weight values and redefines the similarity between data points by incorporating both first-order and second-order shared nearest neighbors. It then computes relative density and relative distance by integrating shared nearest neighbor similarities with natural neighbor weight values. Finally, the algorithm groups samples into clusters. Experiments conducted on six different types of datasets demonstrate that the proposed algorithm notably outperforms three benchmark algorithms.