Evaluating Well Stimulation in Carbonates: Rapid Skin Factor Prediction Using Machine Learning

ABSTRACT Accurate prediction of the poststimulation skin factor ( S ) is vital for optimizing productivity in carbonate reservoirs. Conventional approaches, such as well‐test analysis, are costly and time‐consuming, limiting their use for real‐time decision‐making. This study introduces a data‐driven framework that leverages machine learning (ML) techniques to model the nonlinear relationship between stimulation parameters and skin factor. All nine ML models were trained on a field‐acquired dataset of 80 acidizing stimulation cases from fractured Middle Eastern carbonate reservoirs, incorporating nine routinely measured parameters, including permeability, porosity, reservoir pressure, bottom‐hole temperature, initial skin, depth, acid volume, bottom‐hole pressure, and pumping rate. Model performance was validated using standard evaluation metrics ( R 2 , MSE, AARE%), whereas Shapley Additive Explanations (SHAP) analysis was employed solely for model interpretation and feature importance assessment to ensure consistency with reservoir engineering understanding. Among the evaluated algorithms, the Random Forest model achieved the highest predictive accuracy, with a testing R 2 of 0.993, MSE of 0.0023, and AARE% of 1.41%, whereas the stacked ensemble model showed similarly strong performance. Because the present study focuses on predicting poststimulation skin factor, the workflow is best described as a tool that may support stimulation design by providing quantitative insight into expected treatment outcomes. Demonstrating actual improvements in stimulation design would require additional validation through larger datasets, controlled field applications, or comparative design studies.