AI-Based Prediction of Rock Petrophysical Characteristics from Well Logs

Abstract

This study evaluates and compares five supervised machine learning regression algorithms , Multiple Linear Regression (MLR), Decision Tree Regression (DTR), Random Forest Regression (RFR), Gradient Boosting Regression (GBR), and Extreme Gradient Boosting (XGBoost), for the prediction of three key petrophysical parameters (effective porosity, water saturation, and shale volume) from wireline well log data in the Triassic Argilo-Gréseux Inférieur (TAGI) reservoir of the Berkine Basin, eastern Algeria. The study uses a dataset of 12 wells with 32 measured and interpreted log parameters. XGBoost consistently outperforms competing algorithms, achieving R² values of 0.93–0.96 through optimized 5-fold cross-validation. SHAP-based feature importance analysis confirms the physical consistency of model predictions, identifying gamma ray, resistivity suite, and sonic transit time as the dominant predictors. The methodology provides a validated, reproducible workflow for data-driven reservoir characterization applicable to analogous Triassic sandstone systems in North Africa.