Data driven EM constraints for mixtures offactor analyzers

Mixtures of factor analyzers are becoming more and more popular in the area ofmodel based clustering of high-dimensional data. In this paper we implement a data-driven methodology to maximize the likelihood function in a constrained parameter space, to overcome the well known issue of singularities and to reduce spurious maxima in the EM algorithm. Simulation results and applications to real data show that the problematic convergence of the EM, even more critical when dealing with factor analyzers, can be greatly improved.