A three-way adaptive density peak clustering (3W-ADPC) method

To address the difficulty of determining a clear-cut boundary of a cluster, three-way clustering methods search for a new type of cluster structure characterized by a pair of a core region with tightly connected objects and a fringe region with relatively loosely connected objects. The density peak clustering (DPC) algorithm is a noniterative process that does not require a predetermined number of clusters. In this paper, taking advantage of these two classes of clustering methods, we propose a new three-way adaptive density peak clustering (3W-ADPC) method. The main contribution of proposed method is that the 3W-ADPC algorithm can adaptively select the most applicable neighbor (i.e., the natural nearest neighbor) for each sample and does not need the parameter of a cutoff distance threshold based on two improved definitions of local density and local distance. In other words, 3W-ADPC is a parameter-free three-way clustering algorithm. The experimental results show that the 3W-ADPC algorithm can not only provide an explainable clustering structure, but also has good performances.