On the use of random neural networks for traffic matrix estimation in large-scale IP networks