IMPROVING GRADE ESTIMATION USING MACHINE LEARNING
A COMPARATIVE STUDY OF ORDINARY KRIGING AGAINST MACHINE LEARNING ALGORITHMS
DOI:
https://doi.org/10.17159/Abstract
This study investigates the use of machine learning (ML) algorithms as alternatives to Ordinary Kriging (OK) for grade estimation in a platinum group elements (PGE) deposit. OK, while widely used, relies on variogram modeling and spatial assumptions that may limit its performance in complex geological settings. The research evaluated the performance of ML algorithms—Random Forest (RF), k Nearest Neighbours (kNN), Support Vector Regression (SVR), Decision Trees (DT), and Linear Regression (LR)—through both point and block estimation. The models were compared using validation metrics such as R2, Root Mean Square Error (RMSE), and Mean Absolute Error (MAE), as well as swath plot analyses to assess spatial prediction accuracy. RF and kNN achieved the highest R2 values (0.85 and 0.83, respectively), indicating superior prediction accuracy over OK (R2: 0.76). In block estimation, RF and NN also produced the lowest RMSE values, demonstrating robust generalisation across larger spatial extents. By contrast, LR and SVR underperformed due to their sensitivity to linear assumptions and hyperparameter tuning. These findings show that ML algorithms, particularly RF and NN, offer significant advantages over OK in capturing non-linear relationships and local grade variations, paving the way for more accurate and flexible mineral resource estimation.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 Richard Charles Anson Minnitt, Aniekan
This work is licensed under a Creative Commons Attribution 4.0 International License.
THE INSTITUTE, AS A BODY, IS NOT RESPONSIBLE FOR THE STATEMENTS AND OPINIONS ADVANCED IN ANY OF ITS PUBLICATIONS.
Copyright© 1978 by The Southern African Institute of Mining and Metallurgy. All rights reserved. Multiple copying of the contents of this publication or parts thereof without permission is in breach of copyright, but permission is hereby given for the copying of titles and abstracts of papers and names of authors. Permission to copy illustrations and short extracts from the text of individual contributions is usually given upon written application to the Institute, provided that the source (and where appropriate, the copyright) is acknowledged. Apart from any fair dealing for the purposes of review or criticism under The Copyright Act no. 98, 1978, Section 12, of the Republic of South Africa, a single copy of an article may be supplied by a library for the purposes of research or private study. No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means without the prior permission of the publishers. Multiple copying of the contents of the publication without permission is always illegal.
U.S. Copyright Law applicable to users In the U.S.A.
The appearance of the statement of copyright at the bottom of the first page of an article appearing in this journal indicates that the copyright holder consents to the making of copies of the article for personal or internal use. This consent is given on condition that the copier pays the stated fee for each copy of a paper beyond that permitted by Section 107 or 108 of the U.S. Copyright Law. The fee is to be paid through the Copyright Clearance Center, Inc., Operations Center, P.O. Box 765, Schenectady, New York 12301, U.S.A. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale.
