Multi attribute predictions of lithology and porosity from seismic data over Akaso field, Niger Delta using Machine Learning Regression Analysis

This study employs machine learning-based multi-attribute regression modeling to predict lithology and porosity in the Akaso Field of the Niger Delta Basin using integrated well log and 3D seismic data. A total of 10 seismic attributes—including amplitude envelope, instantaneous phase, frequency, reflection strength, and acoustic impedance—were extracted and statistically analyzed. Petrophysical parameters were computed across six wells using established equations: shale volume (Vsh) ranged from 0.01 to 0.67 v/v, porosity (ϕ) from 0.10 to 0.40 v/v, and water saturation (Sw) from 0.02 to 0.90, confirming excellent reservoir quality in sand unit C. Six regression models (Linear, Ridge, Lasso, SVR, Random Forest, and XGBoost) were trained on normalized datasets with 80:20 train-test split and 10-fold cross-validation. XGBoost outperformed others, achieving R² = 0.91, RMSE = 0.03, and MAE = 0.02 for porosity prediction. Spatial distribution maps generated using XGBoost revealed porosity zones of 20–50% aligned with hydrocarbon-bearing wells (Akos 002, 006, 009, 013, and 012STI) and shale-prone zones (Vsh > 60%) around Akos 004. Forward stepwise regression selected optimal attributes including instantaneous amplitude, filter slices (35/40–45/50 Hz), and first derivatives, yielding Vsh prediction correlation of 0.39 and porosity correlation of 0.32. Multi-linear regression analysis showed slightly lower Vsh correlation (0.23) and higher porosity correlation (0.47), with feature weights confirming the dominance of instantaneous amplitude and low-frequency attributes. Crossplots between P-impedance and porosity, coded by shale volume and water saturation, delineated three lithofacies: hydrocarbon sands (ϕ = 10–20%, Vsh < 0.1), water sands (ϕ = 15–30%), and shales (ϕ > 25%, Vsh > 0.3). Overall, the ML-based regression framework provided accurate, interpretable, and spatially continuous predictions of lithology and porosity in a geologically complex offshore environment. This contributes to enhanced reservoir characterization and supports data-driven decision-making in hydrocarbon field development.