Shale volume estimation using machine learning methods from the southwestern fields of Iran

Shales are a common constituent of many sedimentary formations in well drilling studies, and the accurate determination of shale volume is crucial for understanding reservoir properties and optimizing drilling operations. However, estimating shale volume presents significant challenges, particularly in complex formations. In recent years, machine learning (ML) algorithms have gained prominence for shale volume estimation due to their capability to manage large datasets and complex relationships. This study aims to compare the performance of several advanced ML models—namely, Artificial Neural Networks (ANNs), Bayesian Algorithm (BA), Least Squares Boosting (Lsboost), Linear Regression (LR), Random Forest (RF), and Support Vector Machine (SVM)—for shale volume estimation using well log data. Nine petrophysical log datasets, including SP, RHOZ, PEFZ, NPHI, HLLS, HLLD, HCAL, and DT, were utilized as input features for training the models. The models were evaluated based on performance metrics such as correlation coefficient (R2), average relative error (ARE), root mean square error (RMSE), and mean squared error (MSE). The results highlight that the RF algorithm achieves the highest accuracy and efficiency, with an R2 value of 0.97. Sensitivity analysis further identifies PEFZ and SP as the most influential parameters in shale volume estimation. Finally, model validation was carried out by comparing the estimated shale volume values with actual measurements from the available datasets.