Kamiike, Ryota Tokushima University|Nippon A&L Inc.
least absolute shrinkage and selection operator regression
Statistical 1H nuclear magnetic resonance (NMR) analyses were conducted with ternary copolymer blends. Two out of the three monomers, acrylonitrile, styrene, and α-methylstyrene, were subjected to radical copolymerization to synthesize three kinds of copolymers that were mixed to prepare binary and ternary copolymer blends. The 1H NMR spectral matrix for the copolymers and their blends (explanatory variables) was combined with the blending parameter matrix (objective variables). Cross-validation with the least absolute shrinkage and selection operator regression confirmed that excellent regression models were constructed with a dataset composed of data for eight copolymers and forty-five binary blends; these were used to predict the blending parameters for the binary blends, such as the chemical compositions and mole fractions of the component copolymers. Accordingly, the models were then used to predict the blending parameters for the ternary blends, which resulted in successful and highly accurate predictions. Other regularized regression models, such as Ridge regression and Elastic Net, were also examined.
The Society of Polymer Science, Japan|Springer Nature
polyjour_55_9_967.pdf 3.55 MB