Subsequently, 10,000 times of Y -random validation was performed to assess the chance correlation. Plot ofR _Y² versusQ _Y² is shown in Figure 3d. The average values of R _Y² andQ _Y² for the 10,000Y -random validation are 0.0521 and 0.0038, respectively, which are much lower than the R ² andQ ²_LOO-CV of the developed model. Therefore, the influence of randomness of the dataset itself on the instability of the QSPR model can be ruled out. William plot was used to visualize the developed model’s application domain. Almost all the plots, as depicted in Figure 4e , are within the tolerance of three standard deviations of [-3, 3] and critical leverage level (h ^*=0.1619). It therefore can be concluded that the QSPR model is reliable to predict T _g. The values of relevant statistical parameters of QSPR model are shown inTable 1 .