Determination of polysaccharide content in shiitake mushroom beverage by NIR spectroscopy combined with machine learning: A comparative analysis

A comprehensive comparative analysis of the performance of near-infrared (NIR) spectroscopy combined with various chemometrics and machine learning methods was performed to quantitatively determine polysaccharide content in shiitake mushroom beverages. First, Savitzky-Golay (SG) and orthogonal signal correction (OSC) methods were used for spectral pretreatment. Then, to simplify the model and enhance the generalization performance, variable selection (VS) methods including selectivity ratio (sRatio), variable importance in projection (VIP), recursive partial least square (rPLS), synergy interval partial least square (si-PLS), genetic algorithm (GA), and successive projection algorithm (SPA) were used to select the feature wavelengths. Finally, multiple linear regression (MLR), principal component regression (PCR), partial least squares regression (PLSR), support vector regression (SVR), back propagation neural network (BPNN) and extreme gradient boosting (XGBoost) regression algorithms were used for the multivariate quantitative analysis of NIR data. The results show that different combinations of pretreatment, VS and modelling methods lead to different prediction performances. Except for XGBoost, almost all linear and non-linear models showed satisfying prediction performance with their Rp2 > 0.95. Of these, SG-VS-BPNN models exhibited the best performance with their Rp2 ≥ 0.98. Therefore, NIR combined with machine learning can provide an intelligent, nondestructive and rapid quantitation of polysaccharide content in beverages.