Conformal predictors in chemometric study of mid-infrared food adulteration: quantification of prediction uncertainty

This study introduces a novel application of Conformal Predictor Regression to mid-infrared spectroscopic datasets of adulterated foods. Two datasets were analyzed, and 12 high-accuracy underlying models were tested using normalized inductive nonconformity functions. In all cases, significantly lower margins of error, on average 32.1 % lower, were obtained with normalized nonconformity functions compared to absolute residuals at a 99 % confidence level, without any loss of validity. A significant positive correlation (p < 0.05) was found between the generated error margins and the underlying model's accuracy for all studied cases, and this correlation was independent of the set confidence level. Based on the findings of this study, it is recommended to use a robust solution involving multiple combined normalized Conformal Predictors, with optimal efficiency selected in each case, for quantitative determinations of adulteration using mid-infrared data. Conformal predictors can serve as quantitative estimators of accuracy in vibrational spectroscopy for every individual test sample.