Classification ofABO₃perovskite solids: a machine learning study

Abstract

We explored the use of machine learning methods for classifying whether a particular ABO3 chemistry forms a perovskite or non-perovskite structured solid. Starting with three sets of feature pairs (the tolerance and octahedral factors, the A and B ionic radii relative to the radius of O, and the bond valence distances between the A and B ions from the O atoms), we used machine learning to create a hyper-dimensional partial dependency structure plot using all three feature pairs or any two of them. Doing so increased the accuracy of our predictions by 2-3 percentage points over using any one pair. We also included the Mendeleev numbers of the A and B atoms to this set of feature pairs. Doing this and using the capabilities of our machine learning algorithm, the gradient tree boosting classifier, enabled us to generate a new type of structure plot that has the simplicity of one based on using just the Mendeleev numbers, but with the added advantages of having a higher accuracy and providing a measure of likelihood of the predicted structure.

Theory-Guided Machine Learning in Materials Science

“…For example, in the perovskite formability case above, using a more complicated gradient boosted decision tree reduced the variance in the cross validation accuracy a few percentage points, but did not allow for a simple visual schematic (Pilania et al, 2015). Data transformations that lead to further abstraction like PCA should not be used as inputs to models if there is no clear reason why features should be covariant.…”

“…If one has reason to suspect some functional transformation of the data could help (e.g., for properties determined from constitutive relationships), feature extraction techniques can be used and the results trimmed down by intuition or algorithmically using a technique like LASSO (Ghiringhelli et al, 2015) or feature ranking from an ensemble method like boosted gradient decision trees (Pilania et al, 2015).…”

Stochastic replica voting machine prediction of stable cubic and double perovskite materials and binary alloys

“…Our method enables the prediction of new stable perovskites and the properties of binary compounds. Other works, e.g., [23][24][25] study various aspects of perovskites with existing machine learning algorithms. In the current work, we employed a new and very general machine learning algorithm (whose details will be reported on in [22]) and delineated new phase boundaries in the two classification problems that we in- [19].…”

Compositional optimization of hard-magnetic phases with machine-learning models

“…The objective of this paper is therefore to construct ML models for hard-magnetic properties and to subsequently use them for compositional optimization in order to identify those chemical compositions that exhibit good magnetic properties, but contain only few Re elements. For training the ML models, we select the ThMn 12type crystal structure [31] from our materials database of hard-magnetic phases [32] as a promising substitute for the Re-rich state-of-the-art materials Nd 2 Fe 14 B and SmCo 5 . Instead of using all the data in our database for training, we use a subset with equally-distributed and equally-spaced supporting points where only one alloying element (besides Fe) is considered.…”