Identifying serpentine minerals by their chemical compositions with machine learning

Shichao Ji, Fang Huang, Shaoze Wang, Priyantan Gupta, William Seyfried, Hejia Zhang, Xu Chu, Wentao Cao, J. Zhang Zhou

Research output: Contribution to journalArticlepeer-review

1 Scopus citations


The three main serpentine minerals, chrysotile, lizardite, and antigorite, form in various geological settings and have different chemical compositions and rheological properties. The accurate identification of serpentine minerals is thus of fundamental importance to understanding global geochemical cycles and the tectonic evolution of serpentine-bearing rocks. However, it is challenging to distinguish specific serpentine species solely based on geochemical data obtained by traditional analytical techniques. Here, we apply machine learning approaches to classify serpentine minerals based on their chemical compositions alone. Using the Extreme Gradient Boosting (XGBoost) algorithm, we trained a classifier model (overall accuracy of 87.2%) that is capable of distinguishing between low-temperature (chrysotile and lizardite) and high-temperature (antigorite) serpentines mainly based on their SiO2, NiO, and Al2O3 contents. We also utilized a k-means model to demonstrate that the tectonic environment in which serpentine minerals form correlates with their chemical compositions. Our results obtained by combining these classification and clustering models imply the increase of Al2O3 and SiO2 contents and the decrease of NiO content during the transformation from low- to high-temperature serpentine (i.e., lizardite and chrysotile to antigorite) under greenschist–blueschist conditions. These correlations can be used to constrain mass transfer and the surrounding environments during the subduction of hydrated oceanic crust.

Original languageEnglish (US)
Pages (from-to)315-324
Number of pages10
JournalAmerican Mineralogist
Issue number2
StatePublished - Feb 1 2024
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 2024 De Gruyter. All rights reserved.


  • Serpentine
  • XGBoost
  • classifications
  • clustering
  • k-means
  • machine learning


Dive into the research topics of 'Identifying serpentine minerals by their chemical compositions with machine learning'. Together they form a unique fingerprint.

Cite this