The Prediction of Gold Recovery by Carbon-in-Leach Cyanidation using Visible Near-infrared (VNIR) Spectroscopy of Pulverized Ore Samples from the Cortez Hills Underground Mine

Abstract

A semi-quantitative model was calibrated to predict gold recovery by carbon-in-leach (CIL) cyanidation using visible near-infrared (VNIR) spectroscopy of pulverized ore samples. A model was calibrated using the Cubist rule and instance based regression algorithm on reflectance spectra transformed to the standard normal variate of apparent absorption ( SNV( Log(1/r) ) and reduced to eight components using principal component analysis (PCA). This model, calibrated to the entire dataset has an R-squared value of 0.976 and a root mean squared error of calibration (RMSEC) of 4.87%. Bootstrap validation of the model estimates an R-squared value of 0.813 and a root mean squared error (RMSE) of 13.45% for the population of samples represented by this dataset. Interpretation of the calibrated model suggests that spectral features of iron oxide minerals have a strong influence on gold recovery.