<p>Accurately quantifying irreducible water saturation (Swi) is fundamental to reliable hydrocarbon reserve estimation and effective reservoir management. Traditional empirical correlations often struggle to capture the complex, nonlinear interactions among petrophysical parameters, particularly in heterogeneous formations. To address these shortcomings, this study presents an experimentally validated machine learning framework for Swi prediction. A dataset of 251 core plug samples, incorporating grain density, porosity, permeability, and Nuclear Magnetic Resonance (NMR) T2 logarithmic mean (T2lm), was used to train and evaluate advanced machine learning algorithms. Gradient Boosting Decision Trees (GBDT) optimized through Gaussian Process Optimization (GPO) achieved the highest predictive accuracy, outperforming conventional empirical approaches. SHAP (Shapley Additive Explanations) analysis further enhanced interpretability, identifying permeability as the most affecting factor of Swi. By coupling rigorous laboratory measurements with explainable AI, the framework provides a reliable and scalable methodology for reservoir characterization and hydrocarbon production strategies.</p> Graphical Abstract <p></p> <p>This is a visual summary serves as a pivotal entry point into the research, offering a concise overview of the study’s core findings and methodologies. This study presents a compelling exploration of the successful implementation of data analysis techniques for modeling Swi, serving as a versatile alternative to conventional, time-consuming methodologies. By meticulously integrating essential input parameters such as porosity, permeability, grain density, and Nuclear Magnetic Resonance (NMR), the developed models yield rapid, precise, and scalable predictions that are invaluable for real-time reservoir management. Among the tested algorithms, GBDT emerged as the standout performer, exhibiting exceptional predictive capabilities characterized by a remarkably low Average Absolute Relative Error (AARE%) and a high Coefficient of Determination (R²). These features allow the model to adeptly capture the intricate, non-linear relationships between various stimulation treatments and their resultant effects on reservoir behavior. The robustness of these predictions is supported by a strong correlation between the predicted values and actual field data, substantiated through both cross-plot analyses and evaluations of relative error. Such rigorous validation underscores the reliability and accuracy of the proposed methodologies.Moreover, the integration of SHAP (Shapley Additive Explanations) analysis significantly enhances interpretability, bringing to light reservoir permeability as the foremost influential factor in determining irreducible water saturation. This finding aligns seamlessly with well-established principles in petroleum engineering, as well as empirical observations in the field. The combination of cutting-edge machine learning techniques with foundational domain knowledge not only boosts predictive accuracy but also fortifies the scientific integrity of the findings. Overall, this research illuminates the vast potential of soft computing and explainable AI methodologies to optimize practices related to reservoir management. By facilitating swift, data-driven decision-making processes, the models proposed in this study have the capacity to significantly elevate the efficiency and economic feasibility of hydrocarbon production, particularly in challenging geological formations.</p>