Nonlinear calibration for near-infrared spectroscopy

The nonlinear calibration problem for near-infrared spectroscopy is described. Nonlinear models such as neural networks and support vector machines are used with particular consideration of their generalization abilities as in many real-world examples the calibration data set is sparse. To assess the model predictions without knowledge of the true output, prediction intervals are calculated by the bootstrap, a method known from computational statistics. As an example, the estimation of methyl acetate mole fraction in a four-component mixture is described. The experimental values are taken from a reactive distillation pilot plant.