Prediction of Flight Delays at Beijing Capital International Airport Based on Ensemble Methods

Abstract

Predicting flight delays plays a critical role in reducing financial losses and increasing passenger satisfaction. Due to their ability to combine multiple algorithms, ensemble methods have demonstrated strong predictive performance in many research fields. In this paper, ensemble methods are adopted to predict flight delays. First, based on the current studies, two novel explanatory variables, named arrival/departure pressure and cruise pressure, are proposed as factors affecting flight delays. Second, we introduce the ensemble methods and select representative algorithms for the prediction problem. In addition to the ensemble methods, classical algorithms are also used to predict flight delays. Finally, the actual operational data of Beijing Capital International Airport were utilized to conduct a case study. The results show that the stacking method has better prediction performance than other baseline methods. The mean absolute error (MAE) of the stacking method was about 12.58 min on the test dataset. Furthermore, we tested the effect of the two explanatory variables proposed in this paper, and the results show that the MAE was reduced by about 20% by using the stacking method.