DPC (clustering by fast search and find of density peaks) is a simple and effective density-based clustering algorithm that requires few parameters and does not involve iterative processing. However, it also has limitations, such as sensitivity to the selection of the cutoff distance parameter \(d_{c}\) , and the identification of cluster centers from the decision graph involves subjectivity. Moreover, DPC demonstrates suboptimal performance on datasets with multi-density manifold structures. To address these limitations, a density peaks clustering algorithm based on natural neighbor and multi-cluster merging strategy (DPC-NaN-MS) have been proposed. Firstly, DPC-NaN-MS adaptively identifies the natural neighbor set of each data point and refines local density by incorporating geodesic distance, thereby mitigating the impact of \(d_{c}\) and enhancing clustering performance on datasets with uneven density distributions. Secondly, initial subclusters are formed by searching for natural local density peaks. A novel subcluster merging strategy is introduced, which progressively integrates subclusters until the predefined number of clusters \(k\) is reached. Experimental results on manifold datasets with uneven density distributions and complex morphologies, as well as on real-world datasets, fully demonstrate the effectiveness and superiority of DPC-NaN-MS. Due to the reliance on pairwise distance computation, neighborhood graph construction, and subcluster similarity evaluation, the proposed DPC-NaN-MS algorithm is computationally intensive for large-scale datasets. These operations are inherently parallelizable, making the method well suited for high-performance computing (HPC) and distributed environments. This enables efficient clustering of large-scale, high-dimensional data in real-world applications.
