Machine learning for geochemical exploration: classifying metallogenic fertility in arc magmas and insights into porphyry copper deposit formation

Abstract A current mineral exploration focus is the development of tools to identify magmatic districts predisposed host porphyry copper deposits. In this paper, we train and test four, common, supervised machine learning algorithms: logistic regression, support vector machines, artificial neural networks (ANN) Random Forest classify metallogenic ‘fertility’ in arc magmas based on whole-rock geochemistry. We outline pre-processing steps that can be used mitigate against undesirable characteristics geochemical data (high multicollinearity, sparsity, missing values, class imbalance compositional effects) therefore produce more meaningful results. evaluate classification accuracy each technique using a tenfold cross-validation by testing models deposits unseen during training process. This yields 81–83% for all classifiers, receiver operating characteristic (ROC) curves have mean area under curve (AUC) scores 87–89% indicating probability ranking ‘fertile’ rock higher than an ‘unfertile’ rock. By contrast, bivariate schemes show much lower performance, demonstrating value classifying high dimension space. Principal component analysis suggests porphyry-fertile fractionate deep crust, calc-alkaline associated with Cu-rich porphyries evolve deeper crust alkaline linked Au-rich porphyries. Feature classifiers most important parameters fertile are low Mn, Al, Sr, K listric REE patterns. These signatures further highlight association Cu hydrous undergo amphibole fractionation crust.