Lithology Identification and Estimation of Total Organic Carbon in Organic Shale Through Machine Learning Approaches: Insight From Geochemical Analysis for Source Rock Evaluation

Identification and classification of lithology and estimating total organic carbon (TOC) content in organic shale for source rock evaluation are challenging through indirect approaches in the sedimentary basin and have been addressed in current research through machine learning (ML) approaches. The Kohat sub‐basin is the most prolific basin of Pakistan due to its multiple active petroleum fields and prospective strata ranging from the Cambrian to the Miocene, supported by a hydrocarbon system. While earlier investigations have suggested the potential presence of oil and gas in the source rocks, the region has encountered difficulties making substantial oil discoveries due to a limited understanding of source rock evaluation and complex geological structures. The present study deals with seismic structural interpretation, geochemical analysis for source rock evaluation, lithology identification through ML, and estimation of TOC content using conventional well logs, ML, and lab measured data. The numerical models and ML algorithms based on well log data were applied to estimate TOC content. Lithology delineations through ML were performed within each formation, particularly shale, marl, and limestone in the Patala Formation and sandstone and shale in the Hangu Formation. To evaluate the Paleocene (Hangu and Patala formations) source potential in the basin, a thorough geochemical investigation and source rock evaluation of X‐01 core/well cuttings were conducted. TOC, Rock‐Eval (RE) pyrolysis, vitrinite reflectance techniques, and well log analysis were employed. The TOC values of Hangu Formation are 0.90%–3.20%, which lies in fair to excellent, and Patala Formation 0.82%–2.70%, which shows fair to good TOC content. In this study, it has been inferred that Passey’s method provided better results in estimating the TOC in comparison to core/well cutting measured TOC. The TOC estimate results indicate that the correlation coefficient ( R ) values for well log ∆logR method exceed 0.92 for both formations. In contrast, the random forest (RF)–based ML method demonstrates an R value of 0.94. The Kerogen currently seems to be type II and type III. Generation potential is mostly poor, but at some points, Patala and Hangu show fair to good potential. Study formations’ vitrinite reflectance ( R o ) exists in the oil window. R o values represent vitrinite as the dominant maceral in the Paleocene strata. The second principal maceral is inertinite, and the third maceral is solid bitumen. Pyrite is observed as the main accessory mineral in Paleocene strata. This study proves that well log data can be employed confidently to assess the organic source rock potential even without geochemical data in similar basins around the globe.