Figure
5 The experimental vs. the calculated
(LOO-CV) values for ACGC models of
critical properties: (a) Tc, (b)Pc and (c) Vc.
3.3
Y-randomized analysis
We repeated 1000 Y-randomized validation trials to ensure the
reliability of the results. As was shown in Figure 6, The meanR 2Y andQ 2Y of 1000 Y-random tests of
critical properties model were far lower thanR 2 and Q 2 of the
original model. The comparison of the meanR 2Y andQ 2Y of Y-random test is shown
in Figure 6. This indicates that our model has no chance correlation in
the modeling process.