Figure 4. Results of the QSPR model. (a) plot of calculatedvs. experimental T g of training set and testing set, (b) plot of internal validation via LOO-CV, (c) distribution of errors of the QSPR model and LOO-CV, (d) result of the 10,000Y -randomization tests and (e) William plot.
To further confirm the robustness of the developed model, the scatter plots of the experimental and LOO-CV estimated values, as well as the error distributions for the LOO-CV and the QSPR model, are shown inFigure 4b and Figure 4c , respectively. Specifically,Q 2LOO-CV is 0.8889 and greater than 0.5, showing that the model is robust and stable. Further, the absolute error (AE) distribution of LOO-CV is generally in good agreement with that of the QSPR model, with most polyesters having an error ofT g within 20 ℃.
Table 1. Statistical parameters of the QSPR model