An ensemble-based machine learning solution for imbalanced multiclass dataset during lithology log generation.

Area of Science:

• Geology
• Machine Learning
• Data Science

Background:

• Lithology logs are crucial for understanding subsurface geological structures and correlating drilling data.
• Accurate lithofacies prediction is challenging due to imbalanced multiclass data and geological heterogeneities.

Purpose of the Study:

• To develop a novel workflow for high-resolution lithology log generation using an enhanced weighted average ensemble approach.
• To address the challenges of multiclass imbalanced lithofacies distribution in subsurface geological structures.

Main Methods:

• Implemented a weighted average ensemble approach combined with Error Correcting Output Code (ECOC) and Cost-Sensitive Learning (CSL).
• Utilized machine learning algorithms including Support Vector Machine (SVM), Random Forest (RF), Decision Tree (DT), Logistic Regression (LR), and Extreme Gradient Boosting (XGBoost) as baseline classifiers.
• Trained models on well log data from four wells and evaluated on a blind well from a Middle Eastern oilfield.

Main Results:

• The enhanced weighted average ensemble, particularly based on RF and SVM, demonstrated superior performance in predicting lithofacies.
• Achieved an average Kappa statistic of 84.50% (almost-perfect agreement) and mean F-measures of 91.04% on blind well data.
• The developed workflow proved robust and accurate for high-resolution lithology log production.

Conclusions:

• The novel ensemble-based workflow effectively overcomes challenges associated with imbalanced lithofacies data.
• The enhanced weighted average ensemble significantly improves the accuracy and reliability of lithology log interpretation.
• This approach offers a valuable tool for detailed subsurface geological analysis and resource exploration.