Machine learning-assisted multivariable analysis of glass transition temperature in photopolymerized 4D printed shape memory polymers

Glass transition temperature (T g ) is a critical design parameter for heat-responsive 4D printed shape memory polymers because it defines the programming and recovery temperature window. However, T g in vat photopolymerization printed polymers is controlled by multiple composition and post-curing variables, making exhaustive testing across all possible parameter combinations inefficient and difficult to interpret. A Graeco-Latin square design was used to arrange four parameters, photoinitiator concentration, crosslinker to monomer ratio, post-curing time, and post-curing temperature, at five levels with balanced parameter combinations. Factor contribution analysis showed that post-curing temperature had the largest main effect share at 48.85%, followed by the crosslinker to monomer ratio with 32.83%. In addition, the dataset was used to fit support vector regression (SVR) as the primary predictive model, while linear regression, quadratic response surface model, and Gaussian process regression were included for performance benchmarking. When evaluated using an independent one factor at a time dataset, SVR showed the lowest prediction errors, with an external validation mean absolute error of 1.13 °C, root mean square error of 1.48 °C, and R 2 of 0.89. The SVR model was used to construct multivariable T g response maps and model-derived T g envelopes across the investigated design space. These results demonstrate that a limited and systematically designed experimental dataset can be used for factor contribution analysis, multivariable response mapping and T g prediction in heat-responsive 4D printed polymers.