Predicting Mineral Sites through Machine Learning Model of Buffer Zone Geophysical and Geochemical Trends

The presence of mineral sites could be predicted using a model based on logistic regression. The model is built from geophysical and geochemical property data trends of the buffer zones (areas) around the proven ground truth gold (Au) versus iron (Fe) sites across Australia as the case. The trends summarized within the zones are the mean, minimum value, maximum value, and range of the aforementioned geophysical and geochemical properties. Principal Component Analysis (PCA) is used to increase the quality of data columns (bands) used in building the model, while decreasing the number of columns used. Logistic regression is used in predicting the presence of Au versus iron in individual sites, based on available data on them given by the PCA bands. The plot of PCA bands display diverging directions of trends between Au and Fe sites, while the resulting predictive model possesses an accuracy of 94%.