Density peak clustering algorithm based on adaptive K-nearest neighbors for manifold data

An adaptive K-nearest neighbor clustering algorithm in the light of density will be presented in this paper. When facing manifold structural data, the traditional density peaks clustering algorithm is prone to misclassifying the cluster centers due to the reason of truncation distance, and the use of Euclidean distance can cause misclassification when allocating the remaining samples. The algorithm adaptively defines the K-Nearest Neighbors(KNN) and the extended KNN according to the dataset’s natural distribution, then redefines local density using the KNN and the extended KNN of the data point, at the last uses geodesic distance as the similarity measure to distribute the unallocated samples to the cluster with the highest similarity instead of the cluster center, which effectively avoids sample misclassification. This method can efficiently prevent sample misclassification. Seven algorithms such as DPC algorithm, ADPC - DS have their performance indicators compared with the algorithm in this paper which is applied to manifold structured datasets and real datasets, and the latter has better results.