Multi-Task Local-Global Graph Network for Flight Delay Prediction

Abstract

Airline on-time performance has always been a key factor in evaluating the punctuality of the civil aviation industry and has a profound impact on airlines, airports, and passengers. As a result, there have been increasing demands for the systematic analysis of flight delays and the development of accurate and efficient tools for flight delay prediction. In this paper, a deep learning framework based on graph convolutional networks and multi-task learning is proposed for flight delay prediction. We first use graph convolutional networks to capture the local and global spatial dependencies among the airports. A multi-decoder sequence-to-sequence model is developed to extract the temporal correlation from the data. We further apply a hierarchical graph fusion approach to combine features at different levels of the network to exploit their cross-modality correlations. The model is trained using a dynamic multi-task learning strategy to predict flight arrival and departure delays at the same time to boost the model's generalization and performance. The proposed model is evaluated on a large-scale public flight record dataset against several state-of-the-art methods. The experimental results demonstrate that our model can outperform all baseline methods in predicting short to medium-term flight delays.