Prediction and performance optimization of compressive strength for copper tailings cementitious materials based on stacking ensemble learning

Abstract

[Objective] Due to increasingly severe environmental issues and resource depletion, the sustainable utilization of copper tailings in cementitious materials has attracted widespread attention. However, accurately predicting the compressive strength of cementitious materials incorporating copper tailings remains a challenge due to the complex interactions among material components. A high-precision predictive model was developed using a Stacking ensemble learning approach and optimize mix design to enhance the mechanical properties of the materials. [Methods] Experiments were conducted to investigate the effects of different copper tailings replacement levels(0%, 5%, 10%, 15%, and 20%) and water-to-binder ratios(0.35 and 0.45) on the compressive strength of cementitious materials. To improve the generalization capability of the model, a data fusion method was employed by integrating experimental data with a publicly available concrete compressive strength dataset, [Results] ing in a dataset containing 698 samples. A Stacking ensemble learning model was constructed based on k-nearest neighbors, support vector regression, decision trees, and random forests, with RF serving as the meta-learner. Additionally, Bayesian optimization was applied to fine-tune the hyperparameters of the model to enhance predictive performance. The predictive performance of the Stacking model was evaluated using root mean square error, standard deviation, mean absolute percentage error, and coefficient of determination and was compared with that of individual machine learning models. [Results] The experimental result showed that the compressive strength of cementitious materials generally decreased with increasing copper tailings content, with a significant drop observed when the replacement level exceeded 15%. At the curing age of 28 days, the specimens exhibited the highest compressive strength, indicating a well-developed hydration reaction. The Stacking ensemble learning model demonstrated the best performance in predicting compressive strength, achieving RMSE=0.37, SD=0.16, MAPE=0.91%, and R²=0.991, significantly outperforming individual machine learning models. Among the individual models, RF showed the best performance(RMSE= 2.57, R²=0.977), while KNN exhibited the lowest predictive accuracy(R²=0.967). [Conclusion] A Stacking ensemble learning-based predictive model was developed for the compressive strength of copper tailings cementitious materials and further enhanced its predictive accuracy through Bayesian optimization. The findings indicate that optimizing the water-to-binder ratio and copper tailings content is crucial for improving the mechanical properties of cementitious materials. The proposed Stacking-based predictive model provides reliable data support for mix design optimization, promoting the sustainable application of copper tailings in construction materials.

Keywords

copper tailings / compressive strength / prediction / machine learning / stacking ensemble / influencing factors