Machine Learning Enhanced Geomechanical Characterization and Wellbore Stability Assessment in the Offshore South Tano Field, Ghana

ABSTRACT: Geomechanical characterization is fundamental to optimizing drilling operations and ensuring wellbore stability in offshore petroleum fields. This study presents a machine learning enhanced geomechanical assessment of the South Tano Field in Ghana’s offshore Tano Basin. Using comprehensive wireline logs (density, porosity, sonic, gamma ray), we calculated vertical stress profiles, analyzed pore pressure distributions, and estimated unconfined compressive strength (UCS) via dual empirical methods (sonic‑based and density‑based). Over the analyzed interval (5,000–6,195 ft), the average vertical stress gradient is 1.13 psi/ft, with overburden stress reaching ~12,000 psi at 12,200 ft depth. Porosity trends indicate probable overpressure onset near 6,200 ft, where deviation from the normal compaction trend would begin; however, the available dataset terminates at 6,195 ft, so this onset is projected rather than directly observed. Traditional UCS estimates from sonic and density logs show systematic discrepancies (mean differences of –6.2 MPa for the sonic method and +5.8 MPa for the density method) across the 5,000–6,200 ft interval. To reconcile these discrepancies, we implemented a comparative machine learning framework including Random Forest Regressor (RFR), XGBoost, LightGBM, Support Vector Regressor (SVR), Artificial Neural Network (ANN), Gaussian Process Regression (GPR), and K‑Nearest Neighbors (KNN). On the independent test set, the Random Forest model achieved excellent predictive performance (R² = 1.00, MAE = 0.045 MPa), substantially improving consistency relative to either empirical method alone. The integrated geomechanical model provides critical inputs for safe mud weight window determination, wellbore trajectory optimization, and reduction of drilling non‑productive time. This study demonstrates the value of combining traditional geomechanical analysis with machine learning for enhanced subsurface characterization in offshore environments.