Highlights

• University of Technology Sydney researchers created an AI platform that predicts leaching efficiency of rare earth elements from secondary resources like mine tailings and e-waste.
• Machine learning models revealed critical factors in metal recovery, with silica concentration and REE classification having the most significant impact on leaching performance.
• The explainable AI approach offers potential for more efficient, economical, and sustainable metal extraction processes across various critical metal recovery streams.

---

With global demand for rare earth elements (REEs) surging toward an estimated USD 10.9 billion market by 2029, researchers led by Quang Loc Nguyen (opens in a new tab) as well as Phong H.N. Vo (opens in a new tab) and colleagues at University of Technology Sydney have unveiled an explainable artificial intelligence (AI) platform capable of predicting—and improving—the leaching efficiency of REEs from secondary resources like mine tailings, industrial by-products, and e-waste. The work promises to make hydrometallurgical recovery cleaner, faster, and more consistent—critical for easing the supply crunch without expanding primary mining footprints.

Study Methods: Machine Learning Meets Metallurgy

The team compiled 572 experimental datasets from the literature, covering chemical compositions, operational parameters, and recovery outcomes. They trained three predictive models—Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and Bayesian Gaussian Process Regression (BGPR)—to forecast leaching efficiency.

The GBM won convincingly with an R² of 0.81 in internal validation, beating BGPR (0.62) and SVM (0.31). Importantly, the researchers paired GBM with SHAP (Shapley Additive Explanations) analysis, making the predictions interpretable in real time and revealing the weight of each variable in driving recovery rates.

Findings

Silica Rules, pH Bites, and Heavy REEs Lag

The AI ranked silica concentration as the single most critical factor. Low-to-moderate silica can aid leaching, but above ~25% it becomes a serious inhibitor, trapping REEs in gel structures. REE classification (light vs. heavy) ranked second—light REEs leach more readily—followed by pH, where lower values consistently boost recovery.

Other factors—aluminum content, temperature, stirring speed, and leaching time—showed smaller but still notable impacts. Interestingly, agitation speed had a non-linear relationship with efficiency, pointing to an optimal range rather than a simple “more is better” rule.

Commercial Relevance

For processors, recyclers, and junior miners looking to monetize tailings or e-waste, the study offers more than academic insight. An AI tool that can flag sub-optimal process conditions—and suggest tweaks in real time—can cut reagent waste, shorten optimization cycles, and raise yields. That’s a direct pathway to improved project economics and ESG credentials.

The approach is also transferable: while tuned here for REEs, the model architecture could be applied to lithium, cobalt, or other critical metals, especially in secondary recovery streams where feed variability is high and lab trialing is expensive.

Limitations & Potential Bias

The authors note a performance drop during external validation, likely due to “domain shift” between training and test data—an inherent risk when aggregating heterogeneous literature data. SHAP, while powerful, assumes additive feature effects and can be misleading if strong non-linear interactions dominate. Also, because data came from published experiments, underrepresented operating conditions may bias the model toward well-studied parameters.

REEx Take

This paper plants a flag in the future of AI-assisted metallurgical design. For investors, the key is that such tools can de-risk secondary recovery ventures, potentially lowering capex tied to pilot plants and shortening time-to-market. The next step—and the commercial litmus test—will be whether these predictions hold up in live circuits, under real-world variability.

Citation: Nguyen, Q.L., et al. “Explainable Artificial Intelligence for Predicting Rare Earth Elements Leaching from Secondary Resources.” Journal of Hazardous Materials, vol. 496, 15 Sept. 2025, 139479. https://doi.org/10.1016/j.jhazmat.2025.139479 (opens in a new tab).

---