From trial-and-error to intelligent prediction: Machine learning drives eco-efficient gold leaching in e-waste

Recovering gold (Au) from e-waste is critical for sustainable resource management, yet traditional optimization methods face challenges such as excessive reagent use, lengthy experimentation, and environmental risks from iterative trials. To address these limitations, we developed a machine learning (ML)-driven framework for intelligent Au leaching optimization. The dataset comprised 11,889 samples with 19 input features, including e-waste properties and leaching process parameters. Four machine learning algorithms were compared to identify the optimal predictive model, among which the XGBoost (XGB) algorithm demonstrated superior predictive performance, achieving R² values of 0.93 (training), 0.86 (testing), and 0.76 (5-fold cross-validation). Feature importance analysis prioritized leaching agent type, concentration, particle size, and leaching time as key determinants of Au leaching efficiency, while partial dependence plot analysis revealed intricate synergies among operational variables. A user-friendly graphical interface was further designed to operationalize these insights, yielding predictions closely aligned with experimental verification results. Life cycle assessments and cost analysis confirmed a 98% reduction in global warming potential (GWP, kg CO₂ eq) and 97% cost savings using this ML-driven predictive approach compared to conventional optimizing methods, underscoring the framework's potential to advance eco-efficient, scalable Au recovery from e-waste and support circular economy objectives.