The MBC Procedure

In some circumstances, the covariance matrix for a Gaussian component can become singular during the progress of the EM algorithm. This terminates the model fitting process. This can happen if the cluster represents a single observation or a set of effectively identical observations. When this happens, PROC MBC issues a note that indicates the existence of the singularity. You might be able to use a different initialization method or random seed to alter the progress of the algorithm and so avoid the singularity, but this is not possible in all cases.