A chemometric approach for the quantitative structural analysis of binary blends of copolymers was conducted. Three types of copolymers were synthesized by radical emulsion copolymerization of two out of three monomers—acrylonitrile, styrene, and α-methylstyrene—to prepare three series of binary blends of these copolymers. Partial least-squares (PLS) regression and least absolute shrinkage and selection operator (LASSO) regression were conducted with datasets in which the 1H nuclear magnetic resonance (NMR) spectral matrix of the binary blends (explanatory variables) is combined with the blending parameter matrix (objective variables) of the binary blends. The blending parameters, such as chemical compositions and mole fractions of the component copolymers, were successfully predicted without any assignments of the 1H NMR signals through stepwise optimization of the objective and explanatory variables. LASSO regression exhibited higher accuracy than PLS regression, suggesting that the variable selection in LASSO regression was responsible for the improvement in the quantitative prediction.
Read full abstract