From database to prediction: Machine learning for 5-f elements coordination using actinide x-ray experimental spectra...

Area of Science:

• Nuclear Chemistry and Materials Science
• Spectroscopy and Computational Modeling

Background:

• The Actinide X-ray Experimental Spectra (AXES) database compiles extensive X-ray absorption spectroscopy (XAS) data for actinides.
• Existing actinide research requires comprehensive spectral datasets and advanced analytical tools for structural property determination.

Purpose of the Study:

• To develop a structural property model for predicting uranium coordination environments using XAS data.
• To identify critical spectral regions influencing coordination number prediction in actinides.

Main Methods:

• Compilation and normalization of experimental XAS spectra within the AXES database.
• Development of a convolutional neural network (CNN) model trained on spectral and structural data.
• Application of Shapley Additive Explanations (SHAP) to interpret model predictions and identify key spectral features.

Main Results:

• A CNN model was successfully trained to predict uranium atom presence in different coordination environments.
• Key spectral regions influencing coordination number prediction were identified: edge and post-edge regions for six-coordination, and edge shape for eight-coordination uranium.
• The model demonstrated potential for enhancing actinide coordination studies.

Conclusions:

• The AXES database and the developed CNN model offer a powerful approach for actinide structural analysis.
• Understanding spectral feature importance advances the predictive capabilities for actinide coordination chemistry.
• Further database expansion and transfer learning can improve model accuracy and reliability.