Data-driven ensemble machine learning for corrosion risk prediction and maintenance optimization in oil and gas pipelines

This research focuses on applying machine learning to predict corrosion risks in oil and gas pipelines, with the goal of enhancing efficiency and reducing downtime. Instead of relying on traditional reactive maintenance, the approach adopts a proactive, risk-based strategy that divides pipeline corrosion into three categories: Low, Moderate, and High. Sections with higher corrosion levels are given priority for immediate maintenance, while lower-risk areas are checked less often, allowing resources to be used more effectively. Ensemble machine learning models was developed and evaluated using key performance metrics, including the coefficient of determination (R²), mean squared error (MSE), root mean squared error (RMSE), and mean absolute error (MAE). Gradient Boosting achieved the best predictive results, with an R² of 0.988, MSE of 1.39 × 10⁻⁶, RMSE of 3.72 × 10⁻⁴, and MAE of 2.17 × 10⁻⁴. Extreme Gradient Boosting, Random Forest, and K-Nearest Neighbours also performed well, whereas the Support Vector Regressor showed much lower predictive capability. The analysis of corrosion risk revealed that most defects fell into the low-risk category, while a significant number were classified as moderate to high risk, emphasizing the need for focused maintenance efforts. Despite the challenges posed by imbalanced datasets, the ensemble approach achieved an overall accuracy of 98.8%, showing consistent and reliable performance across all risk levels. Compared with standard maintenance practices, the proposed framework reduces unnecessary maintenance interventions, improves maintenance prioritization, and demonstrates strong potential for lowering operational costs and downtime, highlighting its effectiveness as a data-driven tool for condition-based pipeline maintenance.