Multivariate Statistical Analysis

In the present study, two multivariate analysis methods, including PLS-DA and PLSR, were carried out. PLS-DA, a widely applied discrimination algorithm, divided the multi-dimensional space into class-regions, hence, the under tested samples were assigned to one specific category. More detailed information can be found in Ruiz-Perez et al. (2020). Three parameters, including the specificity, sensitivity, and accuracy, were calculated in an attempt to estimate the PLS-DA model’s performance. PLSR, a typical regression approach, is commonly utilized for MIR data modeling (Wold et al., 2001). It can be carried out to transform the high-dimensional data into the subspace of latent variables (LVs) through maximizing the covariance of the MIR data with the predicting response variables (Kestens et al., 2008). In the present study, two parameters, including the correlation coefficient of validation (R²p) and root mean square error of validation (RMSEP), were calculated to estimate the PLSR model’s performance. The PLS-DA and PLSR were performed by SIMCA-P+ (Version 13.0, MKS Umetrics) software.