An interpretable and generalised machine learning model for predicting flotation performance

Abstract

Recent advances in sensor technology and machine learning (ML) have created new opportunities to predict and improve flotation performance in mineral processing. However, existing flotation performance prediction models lack interpretability and are mostly poorly generalised, making drawing insights from model outcomes difficult. This study addresses these gaps by developing a balanced flotation model using an Extreme Gradient Boosting (XGBoost) algorithm. To interpret the model, we visualise the decision-making process using a tree plot and analyse the prediction path of the model. We employed SHAP (SHapely Addictive exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to quantify the contribution of each input variable to the model’s predictions. Sensitivity analysis of the input variables revealed patterns that were consistent with the interpretability results, providing additional validation of the model’s decision-making process. The prediction results showed that XGBoost (Extreme Gradient Boosting) model demonstrated superior performance, achieving coefficient of determination ($R^2$) values of 0.97 and 0.92 for training and testing datasets respectively. This performance surpassed comparative models including Gaussian Process Regression (GPR) with $R^2$ values of 0.91 and 0.88 and Support Vector Regression (SVR) yielding the lowest $R^2$ values of 0.76 and 0.75 on training and testing datasets respectively. This study enhances flotation performance prediction while providing clear insights into the model prediction outcomes.