Interpretable machine learning models classify minerals via spectroscopy.

Area of Science:

• Mineralogy
• Geochemistry
• Machine Learning Applications

Background:

• Accurate and rapid mineral identification is crucial across various scientific disciplines.
• Traditional Raman spectral analysis relies on pattern matching, which is challenging for poorly crystalline or mixed-phase environmental samples.
• Existing methods struggle with complex sample matrices and the absence of exact spectral matches in libraries.

Purpose of the Study:

• To develop interpretable machine learning (ML) models for classifying uranium minerals solely from Raman spectral data.
• To enable rapid identification of unknown minerals without requiring an exact spectral library match.
• To create a method that provides a mineral profile of physicochemical properties for unknown samples.

Main Methods:

• Development of interpretable machine learning models trained on Raman spectral data.
• Classification of uranium minerals based on secondary oxyanion chemistry and other physicochemical properties derived from spectra.
• Validation of ML models through correlation with published spectroscopic assignments and classification of novel mineral samples.

Main Results:

• Successfully developed ML models capable of classifying uranium minerals based on Raman spectra.
• The models generate a mineral profile, detailing physical and chemical properties, without direct spectral library matching.
• Model performance was validated by strong correlations with established spectroscopic data and accurate classification of un-trained minerals.

Conclusions:

• The developed ML methodology offers a rapid and confident approach to mineral identification using Raman spectroscopy.
• Physically meaningful classifier models can extract key structural and chemical information from unknown uranium minerals.
• The overall methodology demonstrates broad applicability for the classification of diverse mineral phases.