Initial Seeds Selection in Dynamic Clustering Method Based on Data Depth

Resorting to the theory of atomic models and the tool of data depth, we propose a novel method for initial seeds selection in dynamic clustering method. We define the cohesion of a point in a given data set, which includes the information of the significance and locations of neighboring points together. Then, the dynamic clustering algorithm based on cohesion is proposed. Compared with the density-based dynamic clustering algorithm, the clustering results demonstrate that our proposed method is more effective and robust.