Akpabio, Aniekan2025-07-172024Akpabio, Aniekan . (2024). Improving grade estimation using machine learning: a comparative study of ordinary kriging against machine learning algorithms [Master`s dissertation, University of the Witwatersrand, Johannesburg]. WIReDSpace.https://hdl.handle.net/10539/45601A research report submitted in fulfillment of the requirements for the Master of Science, In the Faculty of Engineering and the Built Environment , School of Mining Engineering, Industrial and Aeronautical Engineering, University of the Witwatersrand, Johannesburg, 2024This study investigated the efficiency of machine learning (ML) methods in the accurate prediction of ore grades, placing them in direct comparison with the established Ordinary Kriging (OK) methodology, a mainstay in geostatistical analysis. Utilising a dataset from a complex platinum group elements (PGE) deposit, the research assessed a suite of ML algorithms—namely, Random Forest (RF), Decision Trees (DT), Support Vector Regression (SVR), and particularly 𝑘- Nearest Neighbours (𝑘NN). The latter is highlighted for its adeptness in assimilating spatial data correlations intrinsically, echoing the insights from Nwalia's analytical explorations. The research engages with detailed swath plot analyses, comparative metric evaluations, and a nuanced understanding of spatial continuity, to illustrate the distinct advantages and operational competencies of the models. 𝑘NN, with its reliance on local data proximities and non-parametric nature, alongside RF, with its ensemble-based approach, emerged as capable in point estimate predictions. These models adeptly delineated local grade variations, demonstrating a high degree of reliability to the observed data and outperforming the OK model in both precision and accuracy. Further, the study examined block estimate predictions, a cornerstone in practical mining and resource estimation, where both 𝑘NN and RF demonstrated a commendable ability to generalise predictions over larger spatial extents. This translates into significant potential for enhancing mineral resource estimation processes, tailoring them to the granular specifics of a given ore body, and refining block model accuracy to inform more strategic mining operations. While the results endorse the ML methodologies as robust alternatives to traditional geostatistical techniques, the research also highlights the nuanced nature of these predictions. Factors such as the ore body's heterogeneity, the appropriateness of the variogram model, and the interplay between prediction scale and algorithmic performance are examined, offering a critical lens through which the suitability of each method is assessed. iv The research suggests that while some models like LR and SVR are bounded by linear assumptions and hyperparameter sensitivities, non-linear models such as DT and RF can innately navigate the complex, multifaceted layers of geological data. The comprehensive evaluation extends to propose a novel set of performance metrics designed to capture the intricacies of grade prediction, thereby aligning closely with the operational demands and decision-making processes in the mining industry.en© 2024 University of the Witwatersrand, Johannesburg. All rights reserved. The copyright in this work vests in the University of the Witwatersrand, Johannesburg. No part of this work may be reproduced or transmitted in any form or by any means, without the prior written permission of University of the Witwatersrand, Johannesburg.UCTDMachine LearningGeostatisticsGrade EstimationDissertationSDG-9: Industry, innovation and infrastructure