Study on Density Peaks Clustering Based on Hierarchical K-Nearest Neighbors

Density Peaks Clustering is a novel clustering algorithm, which can find clusters of arbitrary shapes with fast speed. However, it has a few disadvantages, for example, when the data are unevenly distributed, the clustering performance is not good. Therefore, a improved DPC based on hierarchical k-nearest neighbor (HKNN-DPC) algorithm is proposed, which divided k-nearest neighbors into three layers, each layer of data points has different weight, and redesigned the local density calculation approach. We adopt the proposed algorithm to compare with DPC, DBSCAN and K-means in synthesized and UCI data sets. The experimental results indicated HKNN-DPC had better performance.