Page 74 - Read Online
P. 74
Tang et al. J. Mater. Inf. 2025, 5, 38 https://dx.doi.org/10.20517/jmi.2025.05 Page 15 of 19
Figure 10. UMAP feature analyses on various data subsets of substitutions where the substitution energies are predicted by the CE-SVR
models: (A) Nb, (B) Nb -Nb Si , (C) Nb -Nb Si , (D) Si -Nb Si , (E) Si -Nb Si , and (F) combined Nb Si . UMAP: Uniform manifold
3
II
I
5
5
3
5
3
3
5
I
3
5
II
approximation and projection; CE-SVR: center-environment-support vector regression.
We also explored the functional form of the dependence of the target property on the features, in particular
the roles of nonlinearity and inter-feature coupling. This was possible with the use of GPR-NN hybrid ML
