Enhancing Lithology Classification From Well Logs Using Weighted Class Training and Bayesian Optimization in Voting Classifiers

Document Type : Original Article

Authors

1 Department of Petroleum Engineering, School of Engineering and Engineering Technology (SEET), Federal University of Technology, Owerri P.M.B. 1526, Imo State, Nigeria

2 Shell Nigeria Exploration and Production Company, 21/22 Marina P.M.B. 2418, Lagos State, Nigeria

Abstract

Lithology classification is crucial for understanding subsurface geology and enhancing petroleum resource exploration. This study proposes a voting classifier that combines two base models, namely Weighted Class Random Forest (WCRF) and Bayesian Optimized Extreme Gradient Boosting (XGBoost-BO), to improve lithology classification from well log data. The dataset comprised imbalanced well log data from 10 Norwegian wells in the Utsira formation, totaling 4,327 samples with 7 well logs, and 7 lithology classes. WCRF handled class imbalance by assigning weights to each class based on the reciprocal of class frequencies in the training data, while XGBoost-BO used a balanced training dataset created with the Synthetic Minority Oversampling Technique (SMOTE). The models' performance was assessed using metrics like the F1-Score, the area under the receiver operating characteristic curve (AUC), and confusion matrix. The average AUC values for WCRF and XGBoost-BO were 0.982 and 0.985, respectively, showing high generalization performance. The voting classifier achieved the highest performance, with an average F1-Score of 0.874, surpassing WCRF and XGBoost-BO with F1 scores of 0.866 and 0.861, respectively. This voting classifier enhances the accuracy and efficiency of identifying subsurface rock types, ultimately reducing costs and risks associated with drilling by leveraging data-driven insights.

Keywords

Main Subjects