Comparative study of homogeneous ensemble methods with conventional ML classifiers in litho-facies detection using real-time drilling data

The drilling operation is known to be influenced by the formation’s lithology. Real-time prediction of formation parameters is essential to steer the well and make proper completion decisions—litho-facies aid in quantifying formation qualities, allowing drilling parameters to be optimized. Traditional machine learning (ML) algorithms such as logistic regression (LR) and support vector machine (SVM) are compared to new homogeneous ensemble algorithms such as random forest (RF), adaptive boosting (AdaBoost), and XGBoost for predicting the litho-facies of any formation within the research area in real time. Field data from four wells with over 31,000 data points were used to identify litho-facies in real time. The developed model was trained with fourteen different drilling parameters as independent variables and litho-facies as the dependent variable from the Eagleford shale region of the USA. Drilling parameters such as rotation per minute, rate of penetration, differential pressure, surface torque, gamma-ray correlation, and others are used in the model. The suggested study employs a fivefold cross-validation strategy for developing the models. The F1 and accuracy scores were used to assess the model’s efficiency. When the various machine learning algorithms were examined, it was clear that XGBoost outperformed all other algorithms, with an accuracy of nearly 90%. The proposed approach is unique in the industry since it can predict complicated lithology in real time for vertical/inclined/horizontal wellbores without considering survey parameters.