On the Use of Feed-Forward Neural Networks to Discriminate between Models in Financial and Insurance Risk Frameworks