Improving aircraft ground handling times prediction using machine learning approaches

The growing demand for air transportation places increasing pressure on both airport and airspace capacities, pushing them to their operational limits. As a result, operational optimization and improved traffic predictability have become crucial to efficient resource utilization. Collaborative Decision Making (CDM) procedures have been implemented at major airports to improve coordination among stakeholders and improve traffic predictability. One of the critical milestones in CDM procedures is the Target Off-Block Time (TOBT), which serves as a key indicator to predict the duration of ground handling and potential departure delays. Despite its significance, TOBT predictions are often inaccurate due to unanticipated operational changes during the aircraft turnaround process. This study presents machine learning models based on gradient boosting (CatBoost) and random forest regression methods, which were selected to incorporate ground handling data and improve TOBT prediction accuracy during ground handling operations. Six milestones were defined at which TOBT predictions are refined based on continuously updated data. The results of the case study demonstrate that the prediction accuracy improves progressively as the ground handling data are updated. The overall accuracy improvement is 50.6% compared to the initial accuracy at the first milestone, corresponding to the start of ground handling. The results of this study may encourage airport operators and ground handling companies to implement greater automation in TOBT prediction and updating.