Improved model-based clustering performance using Bayesian initialization averaging