Algorithm for Determining Number of Clusters based on Dichotomy

In current clustering algorithms, the number of clusters must be given in advance. And for a given cluster number K has a great impact on the effect of clustering. However, in practice, the specified cluster number K is mostly determined by experience, while partial use of elbow method to find the optimal cluster number requires high time complexity, and the effect is not always optimal. Therefore, there is no unified and effective method to determine the value of cluster number K. To solve this problem, a method to find the number of clusters is proposed. By dichotomizing the current node data each time, the corresponding contour coefficient is obtained, and the size of the contour coefficient before and after calculation is used to judge whether the clustering needs to continue and whether the current node needs to continue adding children nodes until all nodes cannot continue to generate children nodes, and finally a binary tree is formed. And the final number of leaf nodes n of the generated binary tree is taken as the output of the cluster number of the cluster. The experimental results show that the algorithm performs better in the selection of final cluster number and algorithm time.