Comparison of Response Surface Methodology and Artificial Neural Network for Modeling Xylose-to-Xylitol Bioconversion

Abstract

Previous experimental data of xylose-to-xylitol bioconversion by Debaryomyces hansenii carried out according to a 33full factorial design were used to model this process by two different artificial neural network (ANN) training methods. Models obtained for four responses were compared with those of response surface methodology (RSM). ANN models were shown to be superior to RSM in the predictive capacity, whereas the latter showed better performance in the generalization capability step. RSM with optimization using a genetic algorithm was revealed as a whole to be the best modeling option, which suggests that the comparative performances of RSM and ANN may be a highly problem-specific issue.