A neural network approach for spatial variation assessment – A nepheline syenite case study

Abstract

The present geometallurgical study shows the application of a machine-learning methodology to the prediction of material properties from the Nabbaren nepheline syenite deposit in Norway. The approach used in this study created and tested a shallow neural network along with cluster analysis for the prediction of laboratory concentrate yield and modal mineralogy. The input is bulk chemistry data from the mining company open pit database. The methodology proposed unveils general trends in the deposit to a suitable operational scale for the open pit mine. The accuracy of the prediction models is good, with one of the prediction models achieving a strong correlation coefficient of 0.9. The application of a neural network approach showed a successful attempt in the prediction of concentrate yield and modal mineralogy in the Nabbaren nepheline syenite deposit. However, further investigations in terms of deposit internal variation and mineralogical studies are needed for utilising these prediction models, to further improve the modal mineralogy prediction model by better domaining and for a more representative distribution of samples for modal mineralogy analyses.