Evaluation of Machine Learning Models for Ore Grade Estimation

Abstract

AbstractGeostatistics has been widely used for qualitative estimation of ore deposits for many decades. However, ore quality does not vary uniformly in three dimensions which results in a poor quality estimation with the conventional geostatistical methods. Also, the time required for processing geostatistical data can be substantially high. On the other hand, with the advancement in computational processing power and development of advanced algorithms on artificial intelligence (AI) and machine learning (ML), the requirements of an accurate ore grade estimation in reasonable computation time can be fulfilled. In this paper, the applicability of various machine learning techniques like artificial neural network (ANN), extreme learning machine (ELM), gradient boosted decision trees (GBDT), random forests (RF), support vector regression (SVR) have been discussed for ore grade estimation of different mineral deposits like iron, gold and copper. This study also cross-examines the results of ordinary kriging (OK) and inverse distance weighted (IDW) techniques for qualitative estimation. Correspondingly, statistical parameters such as coefficient of determination (R2) and root mean squared error (RMSE) have also been taken into account for a better understanding of the models. Nowadays, AI/ML techniques are extensively used in multiple fields worldwide, including the mining sector, due to their fast and efficient prediction capability. The investigation of these models highlights the importance of accuracy in predicting the quality of the ore as the latter greatly impacts the economic feasibility of mineral deposits. This study forms a ground for developing new advanced intelligent approaches for improving the accuracy of ore grade estimation for mineral deposits.KeywordsArtificial neural network (ANN)Gradient boosting (GB)Machine learning (ML)Ore grade estimationRandom forest (RF)