A Two-Stage Approach for Flight Departure Delay Forecasting Using Ensemble Learning

Accurate flight departure delay forecasting is essential for reliable travel scheduling in intelligent air transportation systems. A two-stage approach is proposed to classify flight departure delay in the future for airports. We first use a clustering algorithm to set the classification rule according to flight departure delay extracted from history information. In the second stage, several state-of-the-art ensemble learning models, which include random forest (RF), gradient boosting decision tree (GBDT), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), are adopted for the flight departure delay classification. The flight departure delay classification models are trained and validated on flight data collected from Beijing Capital International Airport (PEK). The results show that the LightGBM model performs the best among the four employed models for classifying the flight departure delay. The performance comparison of the models can provide valuable insights for researchers and practitioners.