Figure 21.
The graph shows time in days from 0 to 900 and mean squared error from about 0.1 to 100 degree celsius squared. Curves labelled G C N, GraphSAGE, G A T, Kriging, G P, and A N N rise rapidly at early times, then diverge. Kriging increases steadily to near 100, G C N and GraphSAGE stabilise near 8 to 10, G A T remains near 2 to 4, G P decreases to near 1, and A N N varies between about 0.3 and 5.Comparison of the mean squared errors of the various machine learning algorithms