Integrating nature-inspired optimization techniques and machine learning for accurate CO₂/brine interfacial tension estimation: Implications for CO₂ sequestration and uncertainty analysis

Understanding interfacial behavior between CO₂, in-situ fluids, and porous media is key to ensuring safe containment. However, conventional methods for determining CO₂/brine interfacial tension (IFT) are resource-intensive. To overcome the limitations, we integrated Decision Tree, Support Vector Regression (SVR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) paradigms with nature-inspired optimizers: Particle Swarm Optimization , Gene tic Algorithm, Grey Wolf Optimizer (GWO), and Harris Hawks Optimization , utilizing the most extensive eight-parameter CO₂/brine IFT dataset. We juxtaposed the outstanding model's performance with traditional methods and assessed variable influence via multiple methods. Furthermore, the model was deployed for uncertainty analysis and computation of key parameters that dictate storage integrity. Statistical tests confirmed that model choice had a significant impact on performance, while the effect of the optimizer varied depending on the model. XGBoost achieved the highest predictive accuracy (R² ≥ 0.98), with minimal gains from tuning. It also outperformed existing correlations, achieving R² > 0.94 for diverse CO₂/brine systems, compared to correlations' 0.50–0.90. SVR exhibited the largest performance boost, improving from weak default metrics (R² = 0.80) to competitive results (R² = 0.96) post-tuning. Spearman correlation, permutation importance, and Shapley Additive Explanations identified density difference as the most significant predictor of IFT, while salt type had the least impact. XGBoost–GWO/Monte Carlo simulation yielded P10, P50, and P90 IFT estimates of ∼39, 49, and 63 mN/m, translating to quintessential storage column heights of 4150–6809 m and capacities of 97–159 metric tons/m², demonstrating practical utility. The modeling approach provides an accurate and resource-efficient framework for predicting and optimizing CO₂/brine IFT for safe geo-storage. It could also be applied in early-stage storage site appraisal where limited experimental data are available.