Data-driven prediction of copper leaching yield from brass waste using stacking ensemble learning

The recovery of valuable metals from industrial waste has become increasingly important due to the diminishing supply of primary resources and growing environmental concerns. This paper presents a stacking ensemble learning method for predicting copper leaching yield from brass melting slag under different hydrometallurgical conditions. Instead of conducting time-consuming and complex experiments, a machine-learning approach was built using a large dataset previously collected through controlled laboratory research. Six experimental variables were used as input features, including leaching time, acid concentration, hydrogen peroxide concentration, stirring speed, temperature, and solid-to-liquid ratio. The proposed model combines three tuned base learners, namely Gaussian process regression, least-squares boosting, and support vector regression, with a linear regression meta-learner. The stacking model achieved all individual models in predictive performance, yielding the lowest RMSE of 0.853, MAE of 0.668, and MAPE of 8.347 %, resulting in the highest R2 value of 0.994. The results demonstrate that the proposed method is reliable and practically feasible for predicting leaching yield and optimizing the recovery process of copper from brass waste. Furthermore, it has significant potential for supporting resource management objectives.