Predicting differentiated achondrite parent bodies through machine learning: Insights from major element
Zhao Yan, Jin-Ting Kang, Weibiao Hsu, Fang Huang
Applied Geochemistry
Volume 198, February 2026, 106687
“Highlights
- Machine learning enables accurate classification of achondrites based on major element compositions.
- A fast approach alternative to traditional meteorite classification methods.
- A web interface is provided to enhance accessibility for users.”
“Rapid and accurate identification for parent bodies of achondrites is crucial for planetary science research. In this study, we examined the performance of machine learning algorithms using bulk rock major element compositions to classify the origins of achondrites derived from the Moon, Mars, and asteroid 4 Vesta. Literature data on lunar basalts and anorthosites, Martian meteorites, and HED meteorites potentially originating from Vesta, are compiled and cleaned. Multiple machine-learning models were applied including a Tabular Prior-data Fitted Network (TabPFN) and six classical models including Decision Tree, Gradient Boosting, Support Vector Machine, Random Forest, K-Nearest Neighbors, and Multilayer Perceptron. All models demonstrate robust classification performances achieving over 95 % accuracy for the Test Set. Particularly, the K-Nearest Neighbors and TabPFN models achieve an accuracy exceeding 99 %. This study presents a new, automated method in identifying the parent body of achondrites through bulk rock major element data. While these models perform well, further analysis of feature importance is needed to provide deeper insights into the underlying geochemical controls, ensuring the method complements traditional approaches such as petrography and isotope analysis. To facilitate the broader use by meteorite collectors, cosmochemistry community and enthusiasts, a web interface has been developed to quickly apply this technique: https://geo-cosmo-chemistry.shinyapps.io/meteorites_classification/“
