Journal of Resource Management and Decision Engineering 4 Serial Number 13 2025 12 25 Designing an Integrated Framework Based on Explainable Artificial Intelligence and Data Envelopment Analysis for Evaluating the Performance of Operational Units of the South Zagros Oil and Gas Production Company Designing an Integrated Framework Based on Explainable Artificial Intelligence and Data Envelopment Analysis for Evaluating the Performance of Operational Units of the South Zagros Oil and Gas Production Company 1 21 EN Journal Article 2025 07 08 <p>This study aimed to design and evaluate an integrated framework combining Data Envelopment Analysis (DEA), machine learning, and Explainable Artificial Intelligence (XAI) for assessing, predicting, and interpreting the operational efficiency of oil and gas production units. The study employed an applied quantitative design using 30 decision-making units (DMUs) evaluated for the 2026–2027 operational year. Four inputs comprising feedstock, energy consumption, water consumption, and labor; two desirable outputs comprising main production and revenue; and five undesirable outputs comprising CO₂, NOx, SO₂, PM10, and COD were analyzed. DEA was first used to calculate relative efficiency scores and identify efficient and inefficient units. These efficiency scores were then used as the target variable in supervised machine-learning models, including linear regression, support vector regression, random forest, gradient boosting, and XGBoost. Model performance was evaluated using coefficient of determination, mean absolute error, root mean square error, and cross-validation. SHapley Additive exPlanations (SHAP) were subsequently applied to interpret the selected predictive model and determine the contribution of each operational variable to efficiency. The mean DEA efficiency score was 0.879, indicating an average efficiency gap of 12.1% relative to the estimated frontier. Five DMUs achieved an efficiency score of 1.000 and were classified as efficient. XGBoost demonstrated the best predictive performance, with (R²=0.944), MAE=0.017, and RMSE=0.026; its cross-validated (R²) was 0.889 with RMSE=0.034. SHAP analysis identified energy consumption as the most influential predictor of efficiency (18.6%), followed by main production (16.4%), CO₂ emissions (14.1%), revenue (12.3%), and labor (10.5%). Higher production and revenue generally increased predicted efficiency, whereas excessive energy use, pollutant emissions, and disproportionate labor utilization reduced it. The integrated DEA–machine learning–XAI framework provided a multidimensional, predictive, and interpretable approach to operational performance evaluation.</p> https://journalrmde.com/index.php/jrmde/article/download/441/413