VALIDATION PROCEDURES IN NEAR-INFRARED SPECTROMETRY

Three validation procedures, single evaluation set, cross-validation a nd repeated evaluation set, were tested on near-infrared spectrometric data to evaluate the predictive residual standard deviation and the c omplexity of the regression model based on partial least-squares (PLS) regression. Thirty-six combinations of response variables and predict or variables (originating from three response variables and spectra re corded on the same 60 samples in four laboratories with different inst ruments) were tested. Each validation method was used with several dif ferent percentages of objects in the evaluation sets, from very low pe rcentages (leave-one-out) to 33%. The results show that the frequently used technique of the single evaluation set gives a bad estimate both of the residual standard deviation and of the complexity of PLS model . Cross-validation gives acceptable estimates when at least ten cancel lation groups are used. The validation technique based on the repeated evaluation set, with a large number of repetitions of prediction, giv es excellent estimates of residual standard deviation and of model com plexity, but it requires a very long computing time.