Computational Model Development and Feature Attributes Analysis for Formation Bulk Density: Insights into Sandstone Reservoir Evaluation

Accurate prediction of formation bulk density is a crucial parameter to predict petrophysical and geomechanical properties for hydrocarbon reservoir evaluation and pore pressure prediction for safe drilling operation. When direct surface testing with core analysis is a tedious and time-consuming approach, then bulk density is estimated from a density log, using empirical relationships and a computational model approach. In this paper, the major objectives are to assess the developed log-data driven computational models and use feature attribute analysis to find the most important contributing predictor variables while predicting bulk density for sandstone reservoir rocks. The two widely used ensemble machine learning techniques of bagging-based random forest (RF) and boosting-based extreme gradient boosting (XGB) algorithms for computational models’ development and statistical performance indicators are applied to evaluate model reliability and robustness. An explainable artificial intelligence technique of SHapley Additive exPlanations (SHAP) is utilized to interpret and measure each feature importance for bulk density estimation. The XGB model demonstrates better generalization performance with high determination of coefficient, R-squared (96.36%) and lower statistical error compared to the RF model with R-squared of 92.65%. Considering the single variable elimination process with XGB approach, it is found that formation photoelectric factor and true resistivity are the two most significant feature attributes while gamma-ray is the least contributing feature to obtain bulk density for the studied sandstone gas reservoir. The same relative feature importance ranking is also confirmed by RF and XGB-based SHAP techniques. The proposed modeling strategies and feature scoring can be systematically applied to the precise forecasting of dynamic bulk density profile, porosity and pore pressure prediction for integrated reservoir analysis and hydrocarbon exploration, thereby advancing environmentally sustainable practices while ensuring cost-effectiveness.