An improved weighted multiplicative scatter correction algorithm with the use of variable selection: Application to near-infrared spectra

Abstract Multiplicative light scattering has posed great challenge in near-infrared (NIR) quantitative analysis. When estimating the scattering parameters, uninformative variables for scattering effects may bias the estimation. Weighted least squares (WLS) can be used to avoid the influence of the uninformative variables. In this work, we proposed an improved weighted multiplicative scatter correction algorithm with the use of variable selection (WMSCVS). Baseline is removed first and then variable selection is used to obtain the optimal weights of WLS in estimating multiplicative parameters. The variable selection algorithm, which is designed based on model population analysis (MPA), implements an iterative optimization process. In each iteration, weighted bootstrap sampling (WBS) is used to generate variable subsets and exponentially decreasing function (EDF) is used to control the number of sampled variables. The interpretability and stability of the variable selection results as well as the predictive performance of the corrected spectra were investigated by using two NIR datasets. The experimental results showed that the proposed WMSCVS could give better predictive performance than the state-of-art correction methods.