Cao co-authors paper identifying serpentine minerals using machine learning

Department of Geology and Environmental Sciences Assistant Professor Wentao Cao co-authored a new research paper, “Identifying Serpentine Minerals by their Chemical Compositions with Machine Learning,” published in the journal American Mineralogist.

In the research paper, Dr. Cao and eight other authors address the identification of three serpentine minerals – chrysotile, lizardite and antigorite – using a machine learning approach. Modern techniques to distinguish serpentine minerals typically include transmission electron microscopy and Raman spectroscopy, with their respective advantages and disadvantages.

The authors utilized a new approach – machine learning that is purely based on data – to test the possibility of classifying the serpentine minerals based on their major element variation in SiO2, NiO and Al2O3 and determining crystallization environments. Specifically, the Extreme Gradient Boosting (XGBoost) classification was applied to classify serpentine minerals. A k-means clustering algorithm was employed to test partitioning the 1,375 serpentine minerals from published papers into corresponding geological settings.

The authors demonstrated that high-temperature and low-temperature serpentine minerals show systematically different SiO2 and NiO contents, while similar in MgO, MnO and total iron. Application of the k-means clustering model shows that low-temperature serpentine is distributed into key clusters with corresponding parameters (silhouette scores), and one major cluster is indicative to its geological environments.

American Mineralogist is a scholarly journal of the Mineralogical Society of America.
The paper is currently in press with the following DOI: 10.2138/am-2022-8688. Interested readers can view the paper in the papers in press in the Mineralogical Society of America website online