Machine learning is a trending buzzword in all industries, including mineral exploration. Complex models have been successfully implemented by large multinationals and small start-ups to sell targeted advertising, implement self-driving cars, or perform automatic trading.
In mineral exploration, machine learning models are mainly used for 2D mineral prospectivity analysis and borehole data interpretation. The value of machine learning in these fields is recognised, and both mining and service companies are adding machine learning to their toolbox.
Other promising uses for 3D modelling, ore characterisation or automatic geological mapping from remote sensing data are being developed. Machine learning models have the potential to radically change the way exploration is being conducted from greenfield targeting to resource estimation.
However, algorithms are only as strong as the data that is fed to them. A successful prospectivity map requires extensive expert knowledge to select the right geological, geochemical or geophysical variables that will be used by the algorithm to predict the mineral potential for a specific deposit type. This critical step is often overlooked, but when poor data is fed to an algorithm the predictive model will not provide a satisfying prediction, or worse, mislead the targeting exercise.
Machine learning should be considered a new tool for mineral exploration rather than a replacement for the standard exploration procedure. A state of the art implementation of a prospectivity algorithm on an exploration dataset, backed-up by strong field knowledge and deep deposit expertise can only lead to successful decision making. Similarly, integrating borehole data to automatically interpret lithologies, alteration and structures in boreholes should not replace the logging geologist, but be a support tool for quicker, more accurate and reproducible drill core description and analysis.
Using machine learning improves decision making in exploration and accelerates the data integration and interpretation process. However, algorithms and models should not be used blindly as black boxes. They require upstream and downstream supervision by exploration experts who can critically assess predictions and make the right decision based on them.
