Interpretable Machine Learning Framework Using SHAP for Failure Cause Analysis in Natural Gas Transmission Pipelines

This study develops an interpretable machine learning framework using Shapley additive explanations (SHAP) to analyze gas transmission pipeline failures based on US Pipeline and Hazardous Materials Safety Administration (PHMSA) data (2010–2025). XGBoost achieved optimal performance in classifying six failure modes. SHAP results identified spatial exposure—particularly location type and depth—as dominant risk factors, with right-of-way and shallow-depth segments showing highest vulnerability. Temporal analysis revealed aging effects and seasonal patterns, where natural force damage peaks in winter. Complex nonlinear interactions among operating pressure, pipe diameter, wall thickness, and age underscore the need for integrated safety approaches. These findings support implementing multidimensional integrity management strategies incorporating targeted monitoring and data-driven interventions.