Density Peaks Clustering Algorithm Based on Two-step Allocation Strategy

Density peaks clustering is a simple and efficient density-based clustering algorithm and has attracted wide attention in recent years. However, due to the lack of error correction, the one-step allocation strategy for typical density peaks clustering tends to cause the Domino effect. To solve this problem, this paper proposes a density peaks clustering algorithm based on a two-step allocation strategy. According to the original one-step allocation strategy, we first assign labels to all points. In the first step, we assign the trusted label or the pending label, whether the information of the point and its K-nearest neighbors is consistent, to each point. In the second step, we focus on the trusted labels of the K-nearest neighbors to correct the remaining points with pending labels. Finally, the density peaks clustering algorithm based on the two-step allocation strategy achieves excellent clustering performances on benchmark datasets. The code can be downloaded from https://github.com/Gedanke/DPC-TAS.