Abstract
Accurate prediction of oil–nitrogen interfacial tension (IFT) is critical for designing efficient enhanced oil recovery (EOR) strategies. Traditional empirical correlations often lack generalizability and demand detailed compositional data, motivating the need for robust machine learning frameworks. In this study, Gradient Boosting Machine (GBM) models were developed and optimized using four metaheuristic algorithms including Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO), Coupled Simulated Annealing (CSA), and Whale Optimization Algorithm (WOA) to predict equilibrium IFT under varying pressures, temperatures, and API gravities. A curated dataset of 148 experimental measurements was validated through outlier detection and evaluated using five-fold cross-validation to ensure generalization. Model performance was assessed using R(2), mean squared error (MSE), and average absolute relative error percentage (AARE%). Comparative results demonstrate that the ABC-optimized GBM achieved the highest test R(2) and competitive error metrics, outperforming other optimization strategies in predictive reliability. SHAP analysis further confirmed pressure and temperature as the dominant factors influencing IFT, with API gravity exerting a secondary effect. The findings not only establish the ABC-GBM framework as a powerful predictive tool but also reinforce the physical plausibility of the results, offering practical guidance for process optimization in nitrogen-based EOR applications.