International Journal of Mining and Geo-Engineering

International Journal of Mining and Geo-Engineering

Geological modeling of Kalshaneh-2 coal deposit based on the well logging dataset by machine learning algorithms

Articles in Press

Document Type : Research Paper

Authors
Saeed Soltani
Zahra Aghaei
Maliheh Abbaszadeh
Department of Mining Engineering, University of Kashan, Kashan, Iran.
10.22059/ijmge.2026.403660.595314
Abstract
Automated lithological classification represents a key component of modern coal exploration workflows, especially in settings where core recovery is incomplete and conventional qualitative interpretation of wireline logs remains subjective, operator‑dependent, and time‑intensive. This study targets the development of a robust, one‑dimensional, well‑by‑well machine learning (ML) framework for classifying coal-bearing lithologies, with the ultimate objective of generating reliable, spatially consistent inputs for subsequent three-dimensional geological modeling. The dataset consists of eight wells from the Kalshaneh‑2 coal deposit (West South Khorasan, East Iran), integrating depth‑aligned core‑derived lithology logs with eight conventional wireline measurements systematically available across the field: natural gamma, high‑resolution density, long‑space density, caliper/borehole diameter, long normal resistivity, short normal resistivity, single‑point resistivity, and self-potential. Together, these measurements provide a complementary suite of radiometric, density‑based, resistivity‑based, and electrochemical diagnostics that enhance discrimination among the four lithologies of interest: sandstone, siltstone, shale, and coal. A central component of the methodology is the implementation of a geologically defensible validation protocol. Instead of relying on random sample splitting, the workflow employs a well‑based partitioning strategy. Models were trained and tested on a subset of wells, and subsequently assessed on two completely blind wells. This design enables rigorous evaluation of spatial extrapolation performance, a requirement for practical deployment in mine‑scale modeling. Three ML algorithms, Artificial Neural Network (ANN), Support Vector Machine with a radial basis function kernel (SVM‑RBF), and Random Forest (RF), were comprehensively evaluated. Given the substantial class imbalance in the dataset, with coal constituting only ~7% of all samples, the comparative assessment emphasized per‑class precision, recall, F1‑score, and Cohen’s kappa, rather than relying on overall accuracy, which can be misleading under skewed class distributions. On the internal test set, SVM and RF achieved comparable accuracies of approximately 86%, while ANN reached around 76%. However, in the decisive blind‑well validation, RF exhibited superior generalizability, yielding an accuracy of ~75% (kappa ≈ 0.57), compared with SVM’s ~68% (kappa ≈ 0.46). Notably, RF demonstrated the highest sensitivity to the minority coal class, achieving a recall of ~0.69 and an F1‑score of ~0.61, outperforming SVM (recall ~0.56; F1 ≈ 0.57). These findings highlight the Random Forest algorithm as the most reliable and field‑deployable choice for lithological classification in coal‑bearing sequences under imbalanced data conditions. The proposed 1D ML workflow establishes a technically sound and operationally realistic foundation for downstream stratigraphic correlation, model conditioning, and the construction of high‑resolution 3D geological models for mine planning and resource evaluation.
Keywords
Coal
Lithological classification
Well logging
Random forest
Support vector machine
Artificial neural network
Imbalanced data
Subjects
[1].       Afzal, M., et al. (2022). Machine learning approaches for coal seam identification and lithological classification from well log data. International Journal of Coal Geology.
[2].        Arablouei, A., et al. (2024). Robust machine learning methods for highly imbalanced classification problems in geoscience applications. Journal of Petroleum Science and Engineering.
[3].        Bolour Kavir Mines and Mining Developments Company. (2019). Report on the end of the exploration of Kalshaneh 2 coal mine located in Tabas, South Khorasan Province. Tabas, Iran.
[4].        Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32.
[5].        Cortes, C., & Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3), 273–297.
[6].        Egbe, C., & Emudianughe, J. (2024). Machine learning based lithofacies classification from well log data in heterogeneous reservoirs. Journal of Natural Gas Science and Engineering.
[7].        Farhadi, A., et al. (2022). Well log based lithofacies classification using machine learning with spatially constrained validation. Computers & Geosciences.
[8].         Farhadi, A., et al. (2024). Deep learning-based lithology prediction from well logs under class imbalance conditions. Journal of Petroleum Science and Engineering.
[9].        Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press.
[10].    Hagan, M. T., & Menhaj, M. B. (1994). Training feedforward networks with the Marquardt algorithm. IEEE Transactions on Neural Networks, 5(6), 989–993.
[11].    Hastie, T., Tibshirani, R., & Friedman, J. (2009). The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2nd ed.). Springer.
[12].     Haykin, S. (1999). Neural Networks: A Comprehensive Foundation (2nd ed.). Prentice Hall.
[13].    Karavul, C., et al. (2016). Automated lithofacies prediction from wireline logs using artificial neural networks. Journal of Petroleum Science and Engineering.
[14].     Keskinsezer, A. (2019). Lithofacies modeling and well log interpretation for coal-bearing sequences in the Zonguldak Basin, Turkey. International Journal of Coal Geology.
[15].     Kumar, T., et al. (2022). Evaluation of machine learning algorithms for reservoir facies classification under class imbalance. Journal of Petroleum Exploration and Production Technology.
[16].    Lei, C., et al. (2019). Generalization assessment of machine learning models in subsurface characterization: The role of blind-well validation. Computers & Geosciences.
[17].    Liaw, A., & Wiener, M. (2002). Classification and regression by randomForest. R News, 2(3), 18–22.
[18].    Maiti, S., Tiwari, R. K., & Singh, V. S. (2007). Neural network modeling for lithofacies classification from borehole geophysical logs. Computers & Geosciences, 33(5), 609–621.
[19].    McClelland, J. L., Rumelhart, D. E., & PDP Research Group. (1986). Parallel Distributed Processing: Explorations in the Microstructure of Cognition (Vol. 1). MIT Press.
[20].    Patro, B. (1997). Neural networks and their applications in pattern recognition. IEEE Computer Society Press.
[21].    Raeesi, M., et al. (2012). Reservoir characterization and facies prediction in Iranian coal-bearing formations using well logs. Journal of Asian Earth Sciences.
[22].    Rogers, S. J., Fang, J. H., Karr, C. L., & Stanley, D. A. (1992). Lithofacies classification using artificial neural networks. AAPG Bulletin, 76(5), 731–739.
[23].    Rojas, R. (2013). Neural Networks: A Systematic Introduction. Springer.
[24].    Schölkopf, B., & Smola, A. J. (2002). Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond. MIT Press.
[25].    Vapnik, V. N. (2013). The Nature of Statistical Learning Theory (2nd ed.). Springer.
[26].    Zhang, X., et al. (2024). Mitigating optimism bias in well log-based lithofacies prediction: A blind-well benchmarking study. Marine and Petroleum Geology.
[27].     Zhu, Q., et al. (1999). Support vector machines for lithofacies recognition from well logs. IEEE Transactions on Geoscience and Remote Sensing.
International Journal of Mining and Geo-Engineering

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