Application of Machine Learning Algorithms in Predicting the Performance of Oil and Gas Projects

Oil and gas projects are among the most complex, large-scale, and high-risk industrial endeavors, requiring accurate performance prediction from the early stages. This study aims to predict key performance indicators of such projects—including time, cost, quality, and risk—using machine learning (ML) techniques based on early project characteristics. Data were collected from ۶۷ real-world oil and gas projects in Iran through a structured questionnaire designed and validated with input from industry experts. Key project attributes such as monetary value, duration, and ownership type were used as model inputs. A combination of statistical analysis and various ML algorithms—including Logistic Regression, SVM, Random Forest, KNN, Gradient Boosting, Naive Bayes, and Voting Classifier—was employed to model project performance. Model evaluation was conducted using accuracy (train/test), precision, recall, F۱ score, and accuracy difference as an overfitting indicator. Results revealed that Naive Bayes and Voting Classifier provided the most stable and accurate predictions for cost, quality, and time indicators. Conversely, although Random Forest and Gradient Boosting showed high training accuracy, they suffered from significant overfitting on test data. Logistic Regression also demonstrated balanced and interpretable results, particularly in risk prediction, where other models performed poorly. Overall, the findings suggest that combining machine learning with early-stage project data offers a powerful tool for decision-making, resource allocation, and risk management in oil and gas project environments. This approach introduces a new path for improving performance in complex industrial project management.

کلیدواژه ها:

Project Performance Prediction ، Oil and Gas Projects ، Machine Learning