Improving X-means clustering with MNDL

Estimating the true number of clusters for an unlabeled data set is one of the most important limitations in clustering. To solve this issue, many approaches with different assumptions have been proposed in the literature. X-means clustering is one of the proposed methods, which employs Bayesian Information Criterion (BIC) to approximate the correct number of clusters. In this paper, we propose the use of Minimum Noiseless Description Length (MNDL) as a cluster splitting criterion for X-means clustering. MNDL is able to find the optimum splitting criterion for X-means clustering. Simulation results demonstrate that MNDL splitting criterion has the same computational complexity as BIC but, predicts the true number of clusters more often.