Correlating dynamical mechanical properties with temperature and clay composition of polymer-clay nanocomposites

We propose the development of advanced nonlinear regression models for polymer-clay nanocomposites (PCN) using machine learning techniques such as support vector regression (SVR) and artificial neural networks (ANN). The developed regression models correlate the dynamical mechanical properties of PCN with temperature and clay composition. The input feature space regarding the independent variables is first transformed into high dimensional space for carrying out nonlinear regression. Our investigation shows that the dependence of mechanical properties on temperature and clay composition is a nonlinear phenomenon and that multiple linear regression (MLR) is unable to model it. It has been observed that SVR and ANN exhibits better performance when compared with MLR. Average relative error of SVR on the novel samples is 0.0648, while it is 0.0701 and 7.5909 for ANN and MLR, respectively. The good generalization capability of SVR represents a viable quantitative structure–property relationship (QSPR) model for this dataset across both temperature and clay composition. This better generalization property of a QSPR model is critical concerning practical situations in applied chemistry and materials science. The proposed prediction models could be highly effective in reducing multitude lab testing for developing PCN of desired mechanical properties.