Improving the creditability and reproducibility of variables selected from near infrared spectra

A method based on an assembly of two metrics, including the variable importance in projection (VIP) and the PLS regression coefficients B, was developed for wavelength selection in multivariate calibration of spectral data. The proposed algorithm termed VIP-CARS combined the two metrics in a sequential and iterative manner, rather than directly introducing VIP into CARS-PLS. This approach is particularly attractive for quantification due to its relatively higher reproducibility and robustness compared to the CARS procedure. The method was tested on datasets taken from the corn and Rukuaixiao Tablets. It was shown that a small number of well-defined relevant spectral variables were identified with the proposed approach, providing easy spectral interpretation and high creditability. Moreover, with the implementation of the VIP-CARS algorithm, the prediction performance of the final model and the reproducibility of the selected wavelengths were also improved.