Predicting copper concentrations in acid mine drainage: a comparative analysis of five machine learning techniques

Area of Science:

• Environmental Science
• Geochemistry
• Data Science

Background:

• Acid mine drainage (AMD) poses significant global environmental and health risks.
• Predicting AMD formation is complex due to site-specific variations and limitations of traditional small-scale tests.
• Extrapolating small-scale test results to large mine sites introduces substantial decision-making uncertainty.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting AMD quality.
• To utilize historical monitoring data for creating predictive models.
• To assess the predictive accuracy and uncertainty of various machine learning techniques for AMD.

Main Methods:

• Employed machine learning techniques: Artificial Neural Networks (ANN), Support Vector Machine with polynomial (SVM-Poly) and radial base function (SVM-RBF) kernels, Model Tree (M5P), and K-Nearest Neighbors (K-NN).
• Identified key physico-chemical parameters influencing drainage dynamics as input variables.
• Developed models to predict copper concentrations in AMD using historical site data.

Main Results:

• Evaluated predictive accuracy and uncertainty using statistical measures for all tested techniques.
• Support Vector Machine with polynomial kernel (SVM-Poly) demonstrated the highest predictive performance.
• Performance ranking: SVM-Poly > SVM-RBF > ANN > M5P > K-NN.

Conclusions:

• Machine learning techniques show significant promise for predicting AMD quality.
• Data-driven models can reduce uncertainties associated with traditional AMD prediction methods.
• SVM-Poly is a highly effective tool for forecasting AMD quality, aiding environmental management decisions.