Comment on a recent sensitivity analysis of radial base function and multi-layer feed-forward neural network models

Radial base function (RBF) and multi-layered feed-forward (MLF) networks are two popular types of neural network models. They have recently been compared with respect to their sensitivity to random errors in the input space using Monte Carlo simulations. In this paper it is shown that the major conclusion drawn from the comparison is invalid, i.e. the MLF network is nos more sensitive to random errors than the RBF network for the modeling of the relation between the physical structure and mechanical properties of poly (ethylene terephtalate) yarns. This is a useful result, since the MLF network has the additional advantage of being computationally less expensive. Furthermore, it is shown that theoretical error propagation offers a promising route to the development of prediction intervals for the particular modeling application of interest.