Setting Airport Boarding Strategies Based on Passengers’ Operational Data through Machine Learning Techniques

Abstract

Boarding is crucial to turnaround time and can cause significant delays, with the Federal Aviation Administration (FAA) estimating $30 billion in pre-pandemic losses. Previous studies on airport boarding focus on pre-defined strategies that often overlook passenger behavior. This has led to a lack of consensus on the best way to reduce boarding time and improve the level of service (LoS) in different contexts. To address this, this study proposes modeling boarding time using passenger behavior variables across different strategies by combining different techniques. A simulation of three boarding strategies is conducted using screening design of experiments (DOE) with 24 runs each, resulting in 72 samples for A320 boarding time estimation. Machine learning methods, including linear regression, k-nearest-neighbor (KNN), multi-layer-perceptron (MLP), random forest, and XGBoost, are then applied to the simulation data for analysis. As a result, a model that can be used to predict boarding time for a given context of passenger behavior is discussed. Although random forest and XGBoost showed the highest R-squared values, they presented overfitting. Linear regression, with an R-squared close to 0.5, reveals that boarding strategy and bag distribution are the most influential variables, consistent with the literature. Steffen’s strategy provides the lowest boarding time, averaging 12 ± 0.02 minutes to board 180 passengers.