Optimization of mixture models: Comparison of different strategies (Q1775968)

The main goal of this paper is to compare the performance of several optimisation strategies based on expectation-maximization-type and genetic algorithms, and to determine whether the same algorithm must be used during the entire optimisation phase, or the optimisation should be divided into two stages: an initialisation phase during which optimal starting values are identified, and an optimisation phase with the goal to discover the global maximum. Twenty three different strategies were compared through simulation experiments. The obtained results indicate that two-stage procedures provide the best results, especially when these methods are applied on several sets of initial conditions rather than on one single starting.