Grouped data clustering using a fast mixture-model-based algorithm

Mixture-model-based clustering has become a popular approach in many data analysis problems for its statistical properties and the implementation simplicity of the EM algorithm. However the computation time of the EM algorithm and its variants increases significantly with the sample size. For large data sets, performing clustering on grouped data constitutes an efficient alternative to speed up the algorithms execution time. A rapid and effective algorithm dedicated to grouped data clustering is then proposed in this paper. Inspired by the Classification EM algorithm (CEM), the proposed approach estimates the missing sample at each iteration. An experimental study using simulated data and real acoustic emission data in the context of a flaw detection application on gas tanks reveals good performances of the proposed approach in terms of partitioning precision and computing time.