Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
1563130 | Computational Materials Science | 2009 | 9 Pages |
Atomic properties and ionic conductivity data of perovskite-type oxides were collected from literatures and experiments. The relationship between the electrical conductivity and the atomic property was examined. The oxide ionic conductivities were predicted by using two semi-empirical approaches based on first-principles calculations and three machine learning methods, such as partial least squares (PLS), back propagation artificial neural network (BP-ANN), and support vector regression (SVR). It was found that P/L (the ratio of O–O charge population to the O–O band length) has a quadratic curving relationship with Lnσ (logarithm of oxide ion conductivity) in some undoped perovskite-type oxides. The results of machine learning indicate that the generalization ability of SVR is better than those of BP-ANN and PLS models for predicting Lnσ.