ABSTRACT
The training sets will be used to develop the various calibrations, and the validation sets will be used to evaluate how well the calibrations perform. A calibration can only be as good as the training set which is used to generate it. The training set must be a statistically valid sample of the population comprising all unknown samples on which the calibration will be used. There is an entire discipline of Experimental Design that is devoted to the art and science of determining what should be in a training set. The random approach involves randomly selecting samples throughout the calibration space. The most common random design aims to assemble a training set that contains samples that are uniformly distributed throughout the concentration space. Manual design is most often used to augment a training set initially constructed with the structured or random approach. Chemical and physical nonlinearities are caused by interactions among the components of a system.
