Predicting the porosity and permeability of hydrocarbon reservoirs is a key part of figuring out how much fluid is retained and how much moves through them. However, the absence of conventional porosity logs often complicates these predictions due to factors like borehole instability and logging challenges. This research looks at how well three machine learning algorithms—Linear Regression (LR), Random Forest (RF), and Gradient Boosting (XGBoost)—can predict porosity and permeability from well-log data in the Niger Delta basin. The well-log data includes gamma ray, caliper, density, and compressional sonic logs. The goal of the study was to create and improve machine learning models that could guess the properties of a reservoir without using traditional porosity logs. To get better results, hyperparameters for RF and XGBoost were tweaked, and the models’ work was checked using the coefficient of determination (R²) on training, validation, and blind testing datasets. Results indicated that XGBoost and RF outperformed LR in both porosity and permeability predictions, with R² values reaching 0.94–0.95 for porosity and 0.98–0.99 for permeability in test data. Blind testing further confirmed the robustness of the models, achieving R² values of 0.99 for porosity and 0.999 for permeability. This study adds to what is known in the oil industry by showing how machine learning techniques can be used to accurately predict key reservoir properties. These techniques can be used as a reliable alternative to traditional log data when they are not available, like in the Niger Delta basin. Furthermore, accurate prediction of reservoir properties can optimize operations and reduce uncertainties.

Shallow foundations are a popular and affordable foundation type for construction of buildings and engineering structures. Therefore, precise assessment of the underlying soil structure’s bearing capacity is critical for their successful application. In this study, a multi-deterministic technique has been used to evaluate the bearing capacity of a shallow foundation. Empirical estimation of allowable bearing capacity (BC) was based on data from cone penetration tests employing the Schmertmann’s approach. The BC generally increased with depth across all CPTs, aligning with the observed increase in cone resistance (qc) values. A strengthening soil profile was indicated by the overall rise in the average allowable bearing capacity with depth. The numerical modeling with Plaxis-3Dv24 software application accurately estimated the bearing capacity and settling behavior of the shallow foundation on lateritic clay. The findings are consistent with empirical estimates derived from CPT data, notably for allowable bearing capacity. The average bearing capacity estimated from CPT data was 604.98 KN/m², corresponding to an allowable bearing capacity of 201.66 KN/m². The numerical model predicted an ultimate bearing capacity of 620 KN/m², slightly higher than the empirical estimate, and resulting in an allowable bearing capacity of 206.67 KN/m² with a factor of safety (FoS) of 3 against shear failure. The calculated allowable bearing capacities from both methods are relatively close, indicating a reasonable level of agreement. In terms of settlement, the numerical model predicted initial settlement was 8.0 mm, well within the limiting settlement pressure, while for the empirical data settlement information was available for direct comparison. Therefore, the numerical model provided useful insights regarding settlement. It is critical to recognize that the numerical model’s accuracy is strongly reliant on the input soil parameters (unit weight, Young’s modulus, Poisson’s ratio, cohesion, and friction angle), which were estimated based on field research data and engineering appraisal. Therefore, future research could use advanced constitutive models or laboratory testing to refine these values for more precise numerical simulations. The multi-deterministic technique can be extended to a broader range of case studies involving shallow foundations on lateritic clays, resulting in a more comprehensive database of design variables.