# Data-driven classification of ordinary chondrites and asteroidal metal potential evaluation

## TL;DR

This paper uses machine learning and statistical methods to classify ordinary chondrites and assess their metal content, improving classification accuracy and identifying high-metal asteroids.

## Contribution

A data-driven framework combining ML, PCA, and a new Metal Potential Index for OC classification and asteroidal metal evaluation.

## Key findings

- SVM and RF classifiers achieved 0.90 accuracy in OC classification, outperforming traditional methods.
- Fe/Si and Ni/Si are the most important features for classification, reflecting metal-silicate fractionation.
- The Metal Potential Index ranks H-type chondrites highest for metal content, aiding asteroid resource assessment.

## Abstract

Ordinary chondrites (OCs) comprise ~ 87% of meteorites and are key to decoding early Solar System processes, yet H–L–LL classification is often ambiguous, especially between L and LL. We compile ~ 1100 published bulk analyses and train Support Vector Machine (SVM) and Random Forest (RF) classifiers on 13 geochemical features (including Si-normalized ratios and trace elements). Both models achieve an overall accuracy of 0.90, with precision of 0.96–0.97 (H), 0.86–0.89 (L), and 0.71–0.80 (LL), outperforming traditional single-proxy discrimination. Feature importance is dominated by Fe/Si and Ni/Si, consistent with metal–silicate fractionation trends. Principal component analysis (PCA) shows strong covariance among Fe–Ni–Co and inverse correlations with Si–Mg–Ca, separating metal-rich H from silicate-rich L–LL. To link taxonomy with resource assessment, we propose a Metal Potential Index (MPI) based on (Fe/Si + Ni/Si + Co/Si) after max-normalization; group means decrease from H (1.23) to L (0.87) to LL (0.75). Complementary statistical tests indicate that Fe–Ni (with Co admixtures) show limited dependence on petrographic type within each chemical group, supporting a near-uniform distribution of FeNi (Co) grains at the parent-body scale. Together, ML, PCA, and MPI provide a reproducible, data-driven framework for OC classification and for ranking asteroidal metal potential, identifying H-type parent bodies as the highest-priority metal targets.

The online version contains supplementary material available at 10.1038/s41598-026-35624-0.