There has been a lot of research on flight delays. But it is more useful and difficult to estimate the departure delay time especially three hours before the scheduled time of departure, from which passengers can reasonably plan their travel time and the airline and airport staff can schedule flights more reasonably. In this paper, we develop a Spatio-temporal Graph Dual-Attention Neural Network (SGDAN) to learn the departure delay time for each flight with real-time conditions at three hours before the scheduled time of departure. Specifically, it first models the air traffic network as graph sequences, what is, using a heterogeneous graph to model a flight and its adjacent flights with the same departure or arrival airport in a special time interval, and using a sequence to model the flight and its previous flights that share the same aircraft. The main contributions of this paper are using heterogeneous graph-level attention to learn the influence between the flight and its adjacent flight together with sequence-level attention to learn the influence between the flight and its previous flight in the flight sequence. With aggregating features from the learned influence from both graph-level and sequence-level attention, SGDAN can generate node embedding to estimate the departure delay time. Experiments on a real-world large-scale data set show that SGDAN produces better results than state-of-the-art models in the accurate flight delay time estimation task.
COVID-19 affects aviation around the world. China’s civil aviation almost recovered to its pre-epidemic levels in the domestic market, but there are still local outbreaks that affect air traffic. This paper proposes measuring the impact of local outbreaks of COVID-19 by the machine learning method and the synthetic control method as a counterfactual control group to measure such an impact. In this study, we use the LightGBM algorithm to construct a counterfactual control group and transform the prediction problem from time series to the fitting problem at the spatial level. We find that machine learning methods can measure such an impact more accurately. We take local outbreaks in Beijing and Dalian as examples, and our measure of their impacts shows that the impact of an outbreak on intercity air traffic can be divided into lag, decline, stable, and recovery periods, and will last for a long period (more than 40 days) unless there are external stimuli, such as legal holidays. The outbreaks reduced the number of passengers in the cities by 90%. Finally, we show the impact on the air traffic network, and find that when a local outbreak happens in a big city, tourist cities or small stations will be greatly affected.
10 April 2026 in Transportation Research Part A: Policy and Practice
Accurate flight delay prediction is critical for enhancing passenger satisfaction, optimizing airline resource management, and improving air traffic operations. By anticipating delay occurrences proactively, aviation stakeholders can shift from reactive responses to early interventions, thereby mitigating substantial economic losses and preventing localized flight schedule disruptions from cascading across the broader air transportation network. While deep learning methods have shown promising performance, they are typically trained on fixed datasets with monolithic architectures. Thus, they lack transferability to new flight contexts as they require full retraining to adapt and offer less interpretability into how input features drive delay predictions. To address this limitation, we propose the MoEformer model for flight delay prediction based on a Mixture-of-Experts (MoE) framework. Specifically, MoEformer employs a dynamic sparse gating mechanism that adaptively activates specialized experts for heterogeneous flight characteristics while leveraging shared experts to capture universal delay patterns. This design enables interpretable predictions through expert-feature correlations and supports efficient transferability via modular expansion, where new experts can be added while freezing pre-trained parameters. Experimental results on real-world flight operation datasets demonstrate that the proposed method outperforms state-of-the-art baseline models in departure and arrival delays forecasting. Moreover, it demonstrates strong transferability by maintaining high accuracy on unseen flight contexts without retraining, and provides meaningful interpretability through expert-level feature correlation analysis. This work positions the Mixture-of-Experts paradigm as a promising direction for interpretable and transferable prediction in heterogeneous, real-world environments.
27 February 2026 in Transportation Research Part C: Emerging Technologies