Natural nearest and shared nearest neighbours density peaks clustering algorithm for manifold data

Addressing the challenges faced by the density peaks clustering (DPC) algorithm in precisely identifying cluster centres when dealing with manifold data and its tendency to misclassify samples distant from cluster centres, this paper proposes a natural nearest and shared nearest neighbours' density peaks clustering algorithm for manifold data. The local density is redefined by natural nearest and shared nearest neighbours, which highlights the difference between the cluster centre and other samples, and makes the identification of the cluster centre more accurate. The similarity matrix is constructed according to the similarity between the samples defined in the process of defining the local density to complete the allocation of the remaining samples, which effectively improves the problem of incorrectly allocating samples far away from the centre of the clusters in the manifold clusters. Upon comparing the algorithm described in this paper with DPC and other refined algorithms, the experimental outcomes stemming from the manifold datasets unambiguously demonstrate its proficiency in accurately pinpointing the centroid of each cluster, thus effectively carrying out the clustering task. At the same time, on the UCI datasets and Coil20 dataset, this paper's algorithm can get an ideal clustering effect.