Nonlinear multivariate mapping of chemical data using feed-forward neural networks

Layered, feed-forward neural networks are powerful mathematical tools particularly suited to the mapping of nonlinear, multivariate data. In this paper, the performance of these neural networks, as trained by back-propagation, is compared with some more well-established chemometric techniques, based on matrix regression methods, in the mapping of simulated and experimental data representing a range of different nonlinear multivariate relationships. Results obtained from their studies are consistent with the conclusion that layered, feed-forward neural networks offer increased modeling power as compared with nonlinear, matrix regression, but this higher level of nonlinear modeling performance also has an associated cost