Integrated Machine Learning and Uncertainty Quantification for Predicting Volume of Shale and Lithofacies in the Sub‐Himalayan Fold‐and‐Thrust Belt, Kohat Basin, Pakistan

ABSTRACT Machine learning combined with uncertainty quantification can improve the prediction of shale volume (Vsh) and lithofacies within the Lumshiwal Formation of the Sub‐Himalayan Fold‐and‐Thrust Belt. This study uses a comprehensive suite of open‐hole logs from the Chonai‐01 well to evaluate the performance of supervised learning methods and unsupervised clustering for lithofacies classification and Vsh estimation. The applied algorithms include ensemble‐based models, kernel‐based methods, neural networks, and K ‐Means clustering. Prior to model development, the dataset was carefully preprocessed to standardize log responses, remove outliers, and ensure consistency with geological interpretations. The main contribution of this study is the integration of machine‐learning predictions with uncertainty quantification to improve the reliability of Vsh and lithofacies estimation in a structurally complex fold‐and‐thrust belt setting. The results show that ensemble‐based models, particularly random forest and gradient boosting, consistently outperform other approaches in both classification and regression tasks. These models effectively capture petrophysical contrasts among sandstone, shale, shaly sand, and tight facies, producing high prediction accuracy and stable Vsh estimates with limited uncertainty. In contrast, kernel‐based and neural network models show higher sensitivity near facies boundaries and wider prediction variability, although they still reproduce the overall geological trends. Uncertainty quantification using P10–P50–P90 envelopes for Vsh prediction and confidence curves for lithofacies classification further confirms the robustness of the ensemble models. Two low‐Vsh intervals (2650–2720 and 2750–2830 m) are identified as the most prospective reservoir zones. The Lumshiwal Formation is a key stratigraphic unit in the Kohat Basin, and its heterogeneous shale distribution strongly influences reservoir quality and hydrocarbon potential. Overall, the proposed workflow provides a practical and transferable framework for risk‐aware reservoir characterization in structurally complex clastic systems, with applicability beyond the Kohat Basin to similar fold‐and‐thrust belt reservoirs worldwide.