Machine learning model for multi-factor risk prediction in airport construction under non-stop operations

Airport construction under non-stop operations presents unique safety challenges due to complex multi-factor interactions that traditional qualitative methods cannot adequately address. To address this, a study was conducted on 412 construction events (comprising 103 risk incidents and 309 routine events) at a major international hub airport between 2019 and 2024. First, a risk factor system encompassing six key categories, including personnel, environment, equipment, management, facilities, and operations, was developed, represented by 42 indicator variables. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied. Subsequently, the XGBoost classifier was trained, achieving an accuracy of 92.7%, an Area Under the Curve (AUC) of 0.875, and a recall rate of 85.7%. For model interpretability, SHapley Additive exPlanations (SHAP) values were utilized to quantify feature contributions and elucidate the risk transmission mechanism. Five core risk factors were identified: flight density, visibility, the timeliness of NOTAM release, peak hours, and the experience of construction personnel. Key thresholds were determined: flight density of 35 flights per hour, visibility of 3 km, and a 2-h delay in NOTAM release. SHAP analysis evaluated the synergy of operational pressure variables. These findings provide a foundation for integration into operational risk warning systems, supporting differentiated risk management and offering a data-driven approach to balancing efficiency with construction safety.