p-robust network cost efficiency with genetic algorithms and machine learning

Original data envelopment analysis models for expected cost-efficiency evaluation lack robustness in the presence of uncertainty and high-dimensional data. This gap becomes more critical when dealing with big data in the petroleum industry, where selecting relevant variables from large, noisy datasets significantly affects performance results. To address this gap, we propose an uncertainty-integrated, two-stage network data envelopment analysis framework that incorporates artificial intelligence techniques, genetic algorithm and random forest for optimal feature selection. Genetic algorithm simulates natural selection to identify the most relevant variables, reducing dimensionality and enhancing model stability across probabilistic scenarios. In the second stage, Wilcoxon statistical testing and a p-robust approach are applied to ensure consistent and reliable ranking of decision-making units under uncertain conditions. Random forest complements this framework by capturing hidden data patterns, improving accuracy and interpretability. The model is validated using real-world data from ten oilfields, demonstrating substantial improvements over the traditional data envelopment analysis models in feature selection, expected cost-efficiency measurement, and decision robustness. This study offers a practical and intelligent decision-support tool for expected cost efficiency measurement under uncertainty in complex petroleum environments.

کلیدواژه ها:

two-stage network ، Data envelopment analysis ، Expected cost efficiency ، p-robust ، Wilcoxon Test