Advanced geoscience targeting through focused machine learning

With the majority of the easy mineral resource targets having already been found, the mining industry is being forced to re-evaluate methodologies for conducting exploration programs. Future targets are likely to be deeper, with little or no surface expression, and may be concealed from conventional exploration techniques by overburden, permafrost or other geologically complex environments. In order to locate such targets, a holistic multidisciplinary approach is required to identify trends across multiple characteristic features, including geology, geophysics and geochemistry. Quantifying and correlating georeferenced features from such a wide array of data types can be best handled using machine learning algorithms. These algorithms can be trained on a sub-sample of known mineral deposit locations such that they are able to identify the trends in the multi-parameter data set which are associated with mineralization. Once the relevant trend has been identified, a machine learning algorithm is able to predict new mineralization zones from the full data set. Many different machine learning algorithms exist, each with advantages and disadvantages. Tailoring an algorithm to this specific application requires an understanding of both the data, as well as the theory supporting machine learning. Such a marriage of expertise can lead to vast improvements in performance and a more reliable outcome in predicting future mineralization zones.

Faculty Supervisor:

Dr. Eldad Haber

Student:

Justin Granek

Partner:

NEXT Exploration Inc.

Discipline:

Geography / Geology / Earth science

Sector:

Environmental industry

University:

University of British Columbia