This paper presents a novel computational procedure of an expected maximization (EM) algorithm for multidimensional Gaussian finite mixture models. The main idea behind this approach is replacing the sample mean and covariance by projection depth-based location and scatter in maximization steps of the EM algorithm. This new approach, namely projection based EM algorithm (PEM) enhances the stability and robustness of the algorithm. Further, the efficiency of the PEM is compared with regular-EM (REM) under real and simulated data. The study shows the superiority of the proposed procedure in the context of mixture models by considering the features such as outlyingness, efficiency and robustness.

projection depth, projection median, Gaussian mixture, EM algorithm, robustness.