Chemometric Amylose Modeling and Sample Selection for Global Calibration Using Artificial Neural Networks

Chemometric arnylose modeling for global calibration, using whole grain near infrared transmittance spectra andsample selection, was used in an artificial neural network (ANN), to assess the global and local models generated, based onsamples of newly bred Indica, Japonica and rice. Global sample sets had a wide range of sample variation for amylose content(0 to 25.9%). The local sample set, Japonica sample, had relatively low amylose content and a narrow sample variation(amylose; 12.3% to 21.0%). For sample selection the CENTER algorithm was applied to generate calibration, validation andstop sample sets. Spectral preprocessing was found to reduce the optimum number of partial least squares (PLS) componentsfor amylose content and thus enhance the robustness of the local calibration. The best model was found to be an ANN globalcalibration with spectral preprocessing; the next was a PLS global calibration using standard spectra. These results pose thequestion whether an ANN algorithm with spectral preprocessing could be developed for global and local calibration models orwhether PLS without spectral preprocessing should be developed for global calibration models. We suggest that global calibra-tion models incorporating an ANN may be used as a universal calibration model.