Real-Time Prediction of Formation Permeability Using Drilling Data and Explainable Machine Learning: A Case Study from the Hassi Tarfa Field

Abstract

This study investigates the use of ensemble machine learning models—XGBoost, CatBoost, AdaBoost, and Gradient Boosting—to predict formation permeability from real-time drilling parameters. Using a dataset of 183 samples, CatBoost showed the best performance (R² = 0.837). After feature selection, XGBoost achieved the highest accuracy (R² = 0.9655). SHAP analysis identified Torque, Flow Rate, and Rate of Penetration as key predictors. The results highlight the benefits of feature selection and explainable AI for real-time reservoir characterization.