Permeability and pore pressure prediction from well logs using machine learning: A study in the Niger delta

Machine learning provides a robust method for characterizing reservoirs in the Niger Delta. This study applied machine learning techniques to well log data to predict permeability and pore pressure. Feature selection identified depth, density, velocity, and porosity as critical variables, while resistivity and neutron porosity (NPHI) showed strong correlations with pore pressure (correlation coefficients: 0.5–1.0). Random forest and gradient boosting emerged as the most effective models, achieving R-squared scores above 0.99 for both permeability and pore pressure predictions. This corresponded to a root mean squared error (RMSE) under 20,000, indicating a precise fit between predicted and actual values. Although the Decision Tree model also performed well (R-squared > 0.99), further optimization could improve its RMSE and generalization. These results highlight the potential of machine learning to enhance reservoir characterization and inform decision-making in oil and gas exploration and production. Accurate predictions of reservoir properties can optimize operations and reduce uncertainties. Future work could expand these findings by integrating additional data, such as 3D seismic information, and applying the models to diverse geological settings. This would improve the robustness and transferability of predictions, enabling more comprehensive reservoir analysis.