Classification of ABO3 perovskite solids: a machine learning study.

Area of Science:

• Materials Science
• Solid-State Chemistry
• Computational Materials Science

Background:

• Classifying perovskite and non-perovskite structures is crucial for materials discovery.
• Traditional methods often rely on simplified structural descriptors.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting ABO3 perovskite structures.
• To enhance prediction accuracy by integrating multiple feature sets.

Main Methods:

• Utilized machine learning, specifically gradient tree boosting, for classification.
• Employed feature pairs including tolerance factors, ionic radii, bond valence distances, and Mendeleev numbers.
• Developed hyper-dimensional partial dependency structure plots.

Main Results:

• Combining feature pairs improved prediction accuracy by 2-3% compared to single pairs.
• The gradient tree boosting classifier achieved high accuracy and provided likelihood estimations.
• A novel structure plot integrating multiple features was generated.

Conclusions:

• Machine learning models effectively predict perovskite/non-perovskite structures.
• Integrating diverse chemical features significantly boosts predictive performance.
• The developed method offers a more accurate and informative approach to materials structure prediction.