Dynamic prediction of aircraft turnaround milestone times using a cascaded gradient boosting model for improved airport collaborative decision-making

Accurate prediction of milestone times in aircraft turnaround operations is crucial for enhancing flight on-time performance and airport operational efficiency within the airport collaborative decision-making framework. This study proposed a multi-output gradient boosting regression tree-based model in a cascaded framework to dynamically predict crucial milestone times of aircraft turnaround operations, with predictions continuously updated throughout the operational timeline. A comprehensive feature set, incorporating flight-related attributes and hierarchical information transmission features from preceding predictions, was developed using operational data from a study airport. The results demonstrate the effectiveness of the proposed method with an initial prediction accuracy higher than 80% within ±5 min for the actual turnaround activity times. Prediction performance improves progressively as turnaround operations advance, with over 60% of activities ultimately attaining prediction accuracy above 95% within the same threshold. Feature importance analysis indicates significant differences in feature contributions to different milestones of the ground handling process. This methodology provides stakeholders with actionable insights to support airport collaborative decision-making initiatives, enabling delay minimization and reduced slot wastage.