Estimation of the Number of Clusters based on Simplical Depth
Md. Moshiur Rahman, Md Abdul Masud, Badhan Mazumder · 2020
The estimation of the cluster numbers is one of the most significant open research problem in clustering domain. The clustering accuracy highly depends on the accurate number of clusters. We propose a method based on simplical depth, namely SDM, in this paper to estimate the number of clusters. We use the recursive feature elimination process to select two most significant features from a given dataset. We perform an initial partition on the selected features for exploring the compactness of clusters. We use simplicial depth method to estimate depth values of the data points in each cluster. From the initial partition, the number of clusters with the highest depth value is considered as the estimated number of clusters. To evaluate the performance of proposed method, we use five state-of-the-art methods on both synthetic and real datasets. We observe that the accuracy of cluster number estimation with simplical depth based method is better than others in most of the cases, which is the key element for the robust and compact clustering solution.