A simple method for multivariate calibration with minimization of the prediction relative error

Abstract

A simple method addressing the problem of minimizing the prediction relative error is proposed for multivariate calibration. The method is based on the use of back-propagation artificial neural network (BP-ANN). The regression objective of the simple method is to minimize the prediction relative error by changing the output values of BP-ANN. With both theoretical support and analysis of near infrared spectroscopic data and ultraviolet spectroscopic data, it is demonstrated that the simple method produced lower prediction relative error than partial least squares (PLS), principal component regression (PCR), and BP-ANN methods for the system with a wide content range. In addition, when we consider the value of the root mean square error of prediction (RMSEP), four methods were found to have a similar prediction performance. The simple method can predict low content more accurately for the system with a wide content range.