An interpretable machine learning framework for real-time copper cathode quality prediction in industrial bioleaching–solvent extraction–electrowinning operations

Abstract Real-time prediction of cathode quality in industrial bio-hydrometallurgical copper production is challenging due to complex process interactions and operational variability. Conventional approaches rely on post-production inspection, limiting proactive defect prevention. This study develops an interpretable machine learning framework to classify cathode quality in real time using 343 days of operational data from a commercial tank bioleaching-solvent extraction-electrowinning facility. Four supervised algorithms, decision tree, k-nearest neighbors, support vector machine, and artificial neural network, were evaluated using 15 key process parameters. The decision tree provided the best overall balance between predictive performance and interpretability, achieving 97.08% accuracy, 94.62% precision, 94.62% recall, a 94.61% F1-score, and an area under the receiver operating characteristic curve (AUC) of 0.965. Key drivers were concentrate moisture (12.14%), acidity (8.37%), and oxidation-reduction potential (6.24%). Its transparent rules could support integration with existing programmable logic controller systems. The framework showed potential to reduce off-specification cathode production through real-time process monitoring and decision support.