Model verification of GMM clustering based on signature testing

This paper provides a new model verification approach for Gaussian Mixture Models (GMM) with application in partitional clustering. The proposed method relies on the statistics of the data and model and transforms them into a denser area. The transformed data and model have smaller variation compared to their original versions. Therefore, this data compression can be employed as a signature test for estimating the number of clusters and model verification. Simulation results illustrate the efficiency of the proposed method compared with a similar statistic test in terms of accuracy and robustness for estimating the number of clusters.