Predicting Airport Passenger Arrivals from Flight Schedules: An Explainable Machine Learning Method

Abstract: Accurately predicting passenger arrivals at airports is important for airports, airlines and other stakeholders such as Transportation Security Administration (TSA), as it provides essential insights that directly impact workforce scheduling, resource allocation, and overall passenger flow management. These predictions enable the optimization of staffing levels at security checkpoints, manage congestion, and ensure smoother passenger experiences, and they facilitate optimal gate assignments and on-time departures. However, predicting when passengers will arrive at the airport is challenging due to a range of factors. These include the distance between passengers’ residences and the airport, the variety of ground access modes passengers may choose, passengers’ prior experiences, expected and actual traffic conditions on the way to the airport and even weather conditions. These elements add significant complexity to forecasting passenger arrival times with accuracy. In this study, we aim to address these challenges by examining the relationship between passengers' actual arrival times at airports and scheduled flight departure times, as the latter are published and known. Unlike previous studies that rely heavily on individual itineraries or passenger-level data, this research takes a novel approach by utilizing aggregated flight schedules and seat capacity data to predict overall passenger arrival patterns. This aggregation enables a more scalable and generalized model that can be applied across different airports, even when passenger-specific information is unavailable. To accomplish this, we employ the Light Gradient-Boosting Machine model along with SHapley Additive exPlanations (SHAP) to empirically analyze and explain the temporal relationship between passenger arrivals at security checkpoints and scheduled flight departures across a diverse range of U.S. airports. By leveraging the interpretability of machine learning models, we hope to uncover the timing patterns and decision-making behaviors of passengers as they plan their airport arrivals. Using the SHAP, our findings reveal synchronization patterns between passenger arrival times and upcoming flight departures at the individual airport level. Such synchronization patterns also exhibit variations in seasons, weekdays, and weekends. This broader, airport-wide understanding of passenger arrival behaviors can help practitioners make both short-term adjustments and long-term plans for workforce scheduling and resource management. Additionally, our study paves the way for future research opportunities. In particular, cross-sectional comparisons of temporal dynamics across different airports could provide even deeper insights into variations in passenger behaviors before arriving at the airport, further enhancing the capacity for predictive models to optimize airport operations.