Scholarlink

Interpretability methods are used to ‘shed light’ onto how ‘black-box’ machine learningMachine learning (ML) modelsModel make their predictions. A local interpretation provides an explanation of features’ contribution to an individual ML prediction while global methods (e.g., SHAP) provide a more general understanding of the ML modelsModel. In this paper, interpretability methods have been applied to Gaussian Process Regression ML modelsModel used to predict rougher flotationFlotation copper recoveryCopper recovery. The investigation includes understanding how the combination of plant and chemistry variables affects feature importance and the prediction of rougher copper recoveryCopper recovery. The results show the importance of three key process variables, namely: throughputThroughput, frotherFrothers in specific tank cells, and pulp redox potential. The SHAP summary plot further highlights the linear or nonlinear relationship between the features and the prediction of rougher copper recoveryCopper recovery.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Local and Global Interpretation of Flotation Recovery Predictive Machine Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kirsten Louw,
  • Jixue Liu,
  • Richmond Asamoah

摘要

Interpretability methods are used to ‘shed light’ onto how ‘black-box’ machine learningMachine learning (ML) modelsModel make their predictions. A local interpretation provides an explanation of features’ contribution to an individual ML prediction while global methods (e.g., SHAP) provide a more general understanding of the ML modelsModel. In this paper, interpretability methods have been applied to Gaussian Process Regression ML modelsModel used to predict rougher flotationFlotation copper recoveryCopper recovery. The investigation includes understanding how the combination of plant and chemistry variables affects feature importance and the prediction of rougher copper recoveryCopper recovery. The results show the importance of three key process variables, namely: throughputThroughput, frotherFrothers in specific tank cells, and pulp redox potential. The SHAP summary plot further highlights the linear or nonlinear relationship between the features and the prediction of rougher copper recoveryCopper recovery.