International Journal of Mining and Geo-Engineering

International Journal of Mining and Geo-Engineering

Geometallurgical modeling using graph neural networks

Articles in Press

Document Type : Research Paper

Authors
Mohammad Fahiminia 1
Mirsaleh Mirmohammadi 1
Omid asghari 2
1 School of Mining Engineering, College of Engineering, University of Tehran, Tehran, Iran.
2 Civil and Environmental Engineering Department, School of Mining & Petroleum Engineering, Faculty of Engineering, University of Alberta, Edmonton, Canada.
10.22059/ijmge.2026.415746.595385
Abstract
Accurate prediction of metallurgical recovery in geometallurgical studies requires models that capture spatial continuity and higher-order multivariate interactions among geological and geochemical variables. Conventional machine learning approaches treat each drillhole sample as an independent observation, which restricts their ability to represent the joint influence of grade, mineralogical proxies, alteration patterns, and spatial position-factors that collectively govern recovery behavior rather than acting in isolation. Graph-based methods partially address this limitation by introducing pairwise relationships between samples; however, geometallurgical interactions are inherently group-based, since recovery behavior typically emerges from the combined response of multiple samples sharing similar lithological, geochemical, and spatial attributes. To capture such group-level dependencies, a Hypergraph Neural Network (HGNN) was developed for recovery classification. Unlike conventional graphs, hypergraphs allow a single hyperedge to connect an arbitrary number of nodes, providing a natural representation of higher-order relationships. In the proposed framework, drillhole samples were represented as nodes, and hyperedges were constructed to group samples that share similar feature characteristics and belong to common spatial-geometallurgical domains. This two-part design allowed the model to understand both the similarities in features and the spatial arrangement within the orebody at the same time. FeO, Fe, S, and magnetic susceptibility (MagSus) were used as input variables, and recovery values were classified into three categories: low, medium, and high. The HGNN achieved an overall accuracy of 0.86, outperforming Random Forest (0.81) and Support Vector Machine (0.78), with F1-scores of 0.94 and 0.84 for the low- and high-recovery classes, respectively. Section-based comparisons further showed that the HGNN produced more continuous and spatially coherent geometallurgical domains than Random Forest. These results demonstrate that hypergraph-based learning provides an effective framework for modeling higher-order spatial and multivariate relationships in geometallurgical datasets, with potential implications for improved domain definition and recovery prediction in mine planning.
Keywords
Geometaurgical modelling
Graph neural network
Geostatistical modelling
Machine learning
Subjects
[1]         L. Pérez-Barnuevo, S. Lévesque, and C. Bazin, “Automated recognition of drill core textures: A geometallurgical tool for mineral processing prediction,” Miner. Eng., vol. 118, pp. 87–96, 2018, doi: 10.1016/j.mineng.2017.12.015.
[2]         H. Talebi, L. J. M. Peeters, A. Otto, and R. Tolosana-Delgado, “A Truly Spatial Random Forests Algorithm for Geoscience Data Analysis and Modelling,” Math. Geosci., vol. 54, no. 1, 2022, doi: 10.1007/s11004-021-09946-w.
[3]         M. Reichstein et al., “Deep learning and process understanding for data-driven Earth system science,” Nature, vol. 566, no. 7743, pp. 195–204, 2019, doi: 10.1038/s41586-019-0912-1.
[4]         H. Talebi, L. J. M. Peeters, U. Mueller, R. Tolosana-Delgado, and K. G. van den Boogaart, “Towards Geostatistical Learning for the Geosciences: A Case Study in Improving the Spatial Awareness of Spectral Clustering,” Math. Geosci., vol. 52, no. 8, pp. 1035–1048, Nov. 2020, doi: 10.1007/s11004-020-09867-0.
[5]         W. Chen, Y. Li, B. J. Reich, and Y. Sun, “Deepkriging: Spatially Dependent Deep Neural Networks for Spatial Prediction,” Stat. Sin., vol. 34, no. 1, pp. 291–311, 2024, doi: 10.5705/ss.202021.0277.
[6]         N. Kakhani, M. Mokhtarzade, and M. J. V. Zoej, “Deep learning spatial-spectral classification of remote sensing images by applying morphology-based differential extinction profile (Dep),” Electron., vol. 10, no. 23, Dec. 2021, doi: 10.3390/electronics10232893.
[7]         C. Sothe, A. Gonsamo, J. Arabian, and J. Snider, “Large scale mapping of soil organic carbon concentration with 3D machine learning and satellite observations,” Geoderma, vol. 405, p. 115402, 2022, doi: 10.1016/j.geoderma.2021.115402.
[8]         T. Hengl, M. Nussbaum, M. N. Wright, G. B. M. Heuvelink, and B. Gräler, “Random forest as a generic framework for predictive modeling of spatial and spatio-temporal variables,” PeerJ, vol. 2018, no. 8, 2018, doi: 10.7717/peerj.5518.
[9]         X. Li, Y. Ao, S. Guo, and L. Zhu, “Combining Regression Kriging With Machine Learning Mapping for Spatial Variable Estimation,” IEEE Geosci. Remote Sens. Lett., vol. 17, no. 1, pp. 27–31, 2020, doi: 10.1109/LGRS.2019.2914934.
[10]       H. Meyer, C. Reudenbach, S. Wöllauer, and T. Nauss, “Importance of spatial predictor variable selection in machine learning applications – Moving from data reproduction to spatial prediction,” Ecol. Modell., vol. 411, p. 108815, Nov. 2019, doi: 10.1016/j.ecolmodel.2019.108815.
[11]       H. Meyer, C. Reudenbach, T. Hengl, M. Katurji, and T. Nauss, “Environmental Modelling & Software Improving performance of spatio-temporal machine learning models using forward feature selection and target-oriented validation,” Environ. Model. Softw., vol. 101, pp. 1–9, 2018, doi: 10.1016/j.envsoft.2017.12.001.
[12]       W. Liu and M. J. Pyrcz, “A Spatial Correlation-Based Anomaly Detection Method for Subsurface Modeling,” Math. Geosci., vol. 53, no. 5, pp. 809–822, 2021, doi: 10.1007/s11004-020-09892-z.
[13]       Z. Zhang, Y. Cai, W. Gong, P. Ghamisi, X. Liu, and R. Gloaguen, “Hypergraph Convolutional Subspace Clustering with Multihop Aggregation for Hyperspectral Image,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 15, pp. 676–686, 2022, doi: 10.1109/JSTARS.2021.3136599.
[14]       Y. Feng, H. You, Z. Zhang, R. Ji, and Y. Gao, “Hypergraph neural networks,” 33rd AAAI Conf. Artif. Intell. AAAI 2019, 31st Innov. Appl. Artif. Intell. Conf. IAAI 2019 9th AAAI Symp. Educ. Adv. Artif. Intell. EAAI 2019, pp. 3558–3565, 2019, doi: 10.1609/aaai.v33i01.33013558.
[15]       Y. Gao, S. Ji, X. Han, and Q. Dai, Hypergraph Computation, vol. 40. 2024. doi: 10.1016/j.eng.2024.04.017.
[16]       M. R. Yang and X. J. Xu, “Recent Advances in Hypergraph Neural Networks,” J. Oper. Res. Soc. China, 2025, doi: 10.1007/s40305-025-00630-y.
[17]       B. Rajabinasab and O. Asghari, “Geometallurgical Domaining by Cluster Analysis: Iron Ore Deposit Case Study,” Nat. Resour. Res., vol. 28, no. 3, pp. 665–684, 2019, doi: 10.1007/s11053-018-9411-6.
[18]       M. Pezeshkpor, “Geology and mineralization of the Dardvey iron ore deposit,” 2010.
International Journal of Mining and Geo-Engineering

Articles in Press, Accepted Manuscript
Available Online from 08 September 2026