LAUSR

Abstract For the rapid measurement of an analyte in complex samples using near-infrared (NIR) spectroscopy, appropriate wavelength selection is an important and albeit difficult aspect, which is essential for improving prediction performance. Based on equidistant combination partial least squares (EC-PLS), an equivalence model set was proposed. A wavelength selection method, called repetition rate priority combination PLS (RRPC-PLS), was further proposed and applied for NIR analysis of human serum albumin. The competitive adaptive reweighted sampling combined PLS (CARS-PLS) and EC-PLS, which are well-performed wavelength selection methods, were also conducted for comparison. Based on the various divisions of calibration and prediction sets, the modeling was performed to achieve parameter stability. The posterior sample group excluded in modeling was used to validate and achieve an objective evaluation. Using CARS-PLS, EC-PLS and RRPC-PLS methods, the selected optimal models included 32, 36 and 24 wavelengths, respectively. A simpler and high performance model with 15 wavelengths was also selected with RRPC-PLS method. The root-mean-square errors and correlation coefficients for validation were 0.505 g L −1 and 0.997 for the optimal RRPC-PLS model and 0.530 g L −1 and 0.997 for the RRPC-PLS model ( N =15), respectively. The validation effects were superior to the previous two methods in two aspects of prediction performance and model complexity. The prediction values were close to the measured values with high precision. The results showed that RRPC-PLS is the good improvement on EC-PLS, which can be more effective to enhance the prediction performance and remove the redundant wavelengths.