This paper describes the use of response surface methodology (RSM) to model the performance of a neural network. This is in order to help select the values for the neural network parameters. The method was applied to design a multilayer perceptron network for classifying surface defects on wood veneer. The results show that the performance of the neural network was improved by this method, but extrapolation outside the tested parameter range should be avoided.
PhamD. T.LiuX.Neural Networks for Identification, Prediction and Control, 1997 (Springer-Verlag, London).
2.
RitchieM. D.WhiteB. C.ParkerJ. S.HahnL. W.MooreJ. H.Optimization of neural network architecture using genetic programming improves detection and modelling of gene-gene interactions in studies of human diseases. BMC Bioinformatics, 2003, 4(28), 1–14.
3.
PackianatherM. S.DrakeP. R.Neural networks for classifying images of wood veneer—Part 2. Int. J. Advd Mfg Technol., 2000, 16(6), 424–433.
4.
PackianatherM. S.DrakeP. R.RowlandsH. R.Optimizing the parameters of multilayered feedforward neural networks through Taguchi design of experiments. Quality and Reliability Engng Int., 2000, 16(6), 461–473.
5.
MontgomeryD. C.Design and Analysis of Experiments, 1984 (John Wiley, New York).
6.
AllenT. T.YuL.Low-cost response surface methods from simulation optimization. Quality and Reliability Engng Int., 2002, 18(1), 5–17.
