Highlight

Random forest regressor as used to predict the formation energies of perovskites trained only with the formation energies of spinels and ICSD compounds. The structural features used for ML model were generated using the Center-Environment (CE) model, which takes in the structure file and transform it into a 6 x N-dimensional features, where N is the number of elemental features.

The construction procedure of Center-Environment features:

1. Identify the center atoms in the unit cell

2. Calculate the distances between center atoms and their neighbors (within a given cutoff radius)

3. Obtain the elemental features for center atoms, denoted as X_center (3 x N dimensions)

4. Use the reciprocal of distances as weights, for each center atom, aggregate the elemental features of their neighboring atoms into one features, denoted as X_neighbor (3 x N dimensions)

5. Concatenate X_center  with X_neighbor