With the spread and development of human civilizations there is a growing demand for raw materials and a need for successful mineral exploration. Machine-learning approaches have evolved rapidly and are applied successfully to many data science problems.

In a recent paper, we introduce a new toolbox for ArcGIS that provides the most up-to-date machine-learning algorithms for prospectivity modelling as well as several preprocessing and evaluation tools.

The integration into the ArcGIS platform allows a streamlined workflow for industry, from modelling to ground-truthing and mobile data collection in the field. This is achieved by synchronizing results with mobile apps such as Collector for ArcGIS to use results directly during fieldwork and to add data into the database.

The modelling approach can be knowledge- or data-driven. In the present study, we use purely data-driven models.

Besides algorithms like Support Vector Machines, Logistic Regression and Random Forests, Boosting algorithms were implemented and tested for performance, processing time and robustness on a dataset from the Iberian Pyrite Belt (IPB). The IPB is very well suited for testing the algorithms as there are many known deposits such as the mine of Río Tinto (Fig. 1) that can be used to train the algorithms.

Mine of Río Tinto

All machine-learning algorithms performed very well on the test data, especially the Boosting and Random Forest algorithms. The new tools are a good data-driven alternative to other methods for prospectivity modelling. Boosting algorithms can be easily applied due to only few parameters that need to be tuned and perform well even with only few samples to train on.

The full article is available as open-access publication and can be found here. The Experimental tools are included in a new version of the Spatial Data Modeler ArcSMD5 (https://github.com/gtkfi/ArcSDM) toolbox running on the ArcMap (10.4 and 10.5) and ArcPro.

Autoren: Melanie Brandmeier, Irving Gibran Cabrera Zamora, Vesa Nykänen, Maarit Middleton