Key Takeaways

  • Predicting copper recovery and ore hardness from routine measurements is one of the best-documented ML applications in geometallurgy right now (2024-2026 case studies).
  • Gradient-boosted tree models — XGBoost, LightGBM, Random Forest via scikit-learn — are the dominant technique, not deep learning.
  • Cancha is the closest thing to a turnkey commercial product built specifically for this use case.
  • This is a solid “Yes” — published, reproducible, and increasingly standard practice, not a research curiosity.

TL;DR

Train a gradient-boosted regression model (XGBoost, LightGBM, or Random Forest) on your historical testwork — mineralogy, geochemistry, and geomet variables as inputs, recovery or hardness as the target — either through a purpose-built platform like Cancha or your own Python pipeline.

How Do I Predict Recovery and Hardness With AI?

Recovery and hardness prediction is arguably the most mature ML application on this entire poster, with a real published track record from 2024-2026 case studies rather than vendor claims to take on faith. The setup is standard supervised learning: gather your historical geometallurgical testwork — mineral assemblage from automated mineralogy (TIMA/MLA/QEMSCAN), elemental assays, alteration type, hardness indices — as features, and use measured copper (or molybdenum, gold, silver) recovery, or a hardness parameter like BWI, as the target variable.

The dominant technique in the literature is gradient-boosted trees rather than deep learning — XGBoost, LightGBM, and Random Forest (available through scikit-learn) consistently perform well on this kind of tabular, feature-engineered geological data, and are far easier to interpret than a neural network when you need to explain to a geologist why the model predicts low recovery in a given domain. If you’d rather not build this in-house, Cancha packages exactly this workflow — its analysis module applies machine learning to propose prediction algorithms for recovery, product quality, and throughput directly from your geometallurgical sample data.

What makes this more trustworthy than some of the other emerging categories on this poster: it’s reproducible. Published case studies commonly report R² figures in the 0.85-0.90+ range for recovery prediction from mineralogical/geochemical inputs, which is strong enough that several operations now treat model output as a genuine input to mine planning, not just a research exercise. The usual caveat still applies — model quality depends entirely on how representative your training testwork is of the ore you’ll actually mine.

Try It With Geocluster

Pulling together the actual published benchmarks for recovery/hardness prediction accuracy — and matching them against your own ore body’s characteristics — is exactly the kind of literature-grounded research Geocluster is built to run.