Code collection for machine learning assisted materials screening in the search for novel half-metallic Heusler alloys
This code collection contains scripts that have been used to screen the Materials Project database for novel (nearly) half-metallic full (L21) and inverse (XA) Heusler alloys by employing machine learning-based methods in order to predict the spin-polarization of the density of states at the Fermi level. The models used have been trained on density-functional theory data collected by collaborators. Each step of the materials screening process is performed by the provided scripts, including Data processing, Hyperparameter Optimization, Model Training & Evaluation, Screening Data (The Materials Project) Download & Processing, Predictive Modelin
Resource information
- DOI
- 10.5281/zenodo.10075055
- Credit basis
- Creators in DataCite DOI metadata
- Publisher
- Zenodo
- Upstream last modified
- Nov 5, 2024
- Access
- Files remain with the original publisher
- Last checked
- Nov 5, 2026
- License
- MIT License
- Last record update
- Nov 5, 2026
About this resource
This code collection contains scripts that have been used to screen the Materials Project database for novel (nearly) half-metallic full (L21) and inverse (XA) Heusler alloys by employing machine learning-based methods in order to predict the spin-polarization of the density of states at the Fermi level. The models used have been trained on density-functional theory data collected by collaborators. Each step of the materials screening process is performed by the provided scripts, including Data processing, Hyperparameter Optimization, Model Training & Evaluation, Screening Data (The Materials Project) Download & Processing, Predictive Modeling, and Filtering of the Predictions. Also, scripts compiling additional explorative visualizations and SHAP-guided explainable artificial intelligence visualizations are included. Upon investigation of the predicted compounds, FLAPW electronic structure validation computations have been performed. Calculation submission, spin-polarization computation and visualization scripts depicting the validation process are included in this publication as well. The XGBoost model which has been used by us is included (stored using the Python package pickle) for future reproducibility of our results. This work was performed as part of the Helmholtz School for Data Science in Life, Earth and Energy (HDS-LEE) and received funding from the Helmholtz Association of German Research Centres.