Method for Deriving Multi-factor Models for Predicting Airport Delays




Xu, Ning

Journal Title

Journal ISSN

Volume Title



Traffic Flow Manage ament (TFM), in coordination with Airline Operation Centers (AOC), manage the arrival and departure flow of aircraft at the nations airports based on the airport Arrival and Departure rates for each 15 minute segment throughout the day. The management of traffic flow has become so efficient in the U.S., that approximately 95% of the delays now occur at the airports (not airborne). Inefficiencies in the traffic flow occur when non-traffic flow delays (e.g. carrier, turn-around, aircraft swapping and non-terminal area weather) are super-imposed on the traffic flow delays. Researchers have correlated these non-traffic flow delays at airports with sets of causal factors and have created models to predict aggregate delays at airports on the time scale of a day. To be consistent with the way traffic flow is managed, a model of causal factors of delays in 15 minute segments would provide the analytical basis for improving the efficiency of TFM. This dissertation describes the development of multi-factor models for predicting airport delays in 15 minute segments at 34 OEP airports. The models are created using Multivariate Adaptive Regression Splines (MARS). The models, generated using historic individual airport data, exhibit an accuracy of 5.3 minutes for generated delay across all the airports, and 2.1 minutes for absorbed delay across all the airports. A summary of the factors that drive the performance of each airport is provided. The sensitivity of each of the factors is also analyzed. Analysis of the models indicates that the factors that determine Airport Delays in 15 minute segments are unique to each airport. The most significant factors that generate delays at most of the nation's airports are Carrier Delay, GDP Delay at the outbound destination, and Departure Demand Ratio. Because of the relationship between these factors, and the propagation of delays throughout the network, the only way to mitigate system-wide delays is via a holistic network approach. The implications of these results are discussed. The potential benefits from this research include providing: (1) researchers and analysts a method to identify systemic causes of delays in the NAS and study the trends of influential factors; and (2) airlines and Air Traffic managers a means to evaluate predicted delays while executing Traffic Flow Management initiatives.



Prediction, Regression, Model, Non-linear, Airport, Delay