A comparison of neural network models for prediction of RF propagation loss

Neural networks have previously been shown to be a powerful tool when used for the prediction of RF propagation loss. In this paper, we present a comparison of the performance of propagation models based on multilayer perceptron networks, radial basis function networks, and conventional multiple linear regression techniques. The complexity of the neural prediction models is set based on the cross-validation principle, with the goal of achieving good generalization properties. A number of methods for selection of training examples are also compared. It is shown that, in most cases, neural prediction models can provide an improvement in performance over conventional empirical models