A Graph-based Approach to Estimating the Number of Clusters

We consider the problem of estimating the number of clusters (k) in a dataset. We propose a non-parametric approach to the problem that is based on maximizing a statistic constructed from similarity graphs. This graph-based statistic is a robust summary measure of the similarity information among observations and is applicable even if the number of dimensions or number of clusters is possibly large. The approach is straightforward to implement, computationally fast, and can be paired with any kind of clustering technique. Asymptotic theory is developed to establish the selection consistency of the proposed approach. Simulation studies demonstrate that the graph-based statistic outperforms existing methods for estimating k. We illustrate its utility on a high-dimensional image dataset and RNA-seq dataset.