This paper presents a data-driven study of wait time patterns for international arriving passengers across all sixty-one terminals from the forty-four airports of entry of the United States. Each airport is an independent entity which operates with various airlines and handles demand volumes differently. This induces seasonal variation in service quality from one airport to another. Exploring six years worth of data, this paper investigates the current and long-term performance trends - an increasing number of flights versus a decreasing number of customs booths - of all airports of entry from a passenger perspective. A performance analysis is then conducted that compares average wait times of incoming passengers, considering incoming traffic ratios and allocated resources. Leveraging machine learning algorithms, six regression algorithms are trained and tested to accurately predict passenger wait times through customs at selected airports. An analysis of the performance of these models shows that the best approach - using a Gradient Boosting regressor for each terminal of entry - can capture the daily and seasonal variations of traffic patterns and immigration booth availabilities with a mean absolute error of less or equal to 5 minutes for twenty-eight terminals of entry and less than 10 minutes for all terminals. Observations show significant disparities across airports that may be explained by the foreign/US passenger ratio and the quality of booth management.
|Original language||English (US)|
|Title of host publication||AIAA/IEEE Digital Avionics Systems Conference - Proceedings|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|State||Published - Sep 1 2019|