Evaluation of machine learning algorithms for grade estimation using GRNN & SVR

Abstract

Estimates of the ore grade need precise location prediction using the limited drill data available. It is a significant and important phase in the process of deciding how to invest in and develop various mining operations. This study analyses the effectiveness of using General Regression Neural Networks (GRNN) and Support Vector Regression (SVR) to improve grade estimate prediction in western India. The algorithm used to estimate grade took into account supplemental lithological information that differed throughout the study area. In this particular investigation, the factors considered as input variables were three-dimensional geographical coordinates and four underlying lithological units. On the other hand, the factors considered as output variables were CaO, Al2O3, Fe2O3, and SiO2. Ordinary kriging (OK), a geo-statistical technique, was utilized in order to verify the results of the comparative analysis performed on these models (OK). The GRNN model performed significantly better than the SVR model and the OK model when it came to generalization and the ability to predict grades.