Integrated Predictive Model of Air Traffic Flow Using Deep Learning

Predicting airport traffic flow accurately is essential for improving air traffic management, minimizing delays, and enhancing operational efficiency throughout the aviation network. This research implements two deep-learning models to predict air traffic flow. A Long Short-Term Memory (LSTM) model is applied to forecast airport traffic flow at an individual airport, focusing on temporal patterns. Additionally, a hybrid deep learning approach that combines Graph Convolutional Networks (GCNs) with LSTM is developed to capture both spatial and temporal dependencies across the airport network, allowing for predictions of traffic flow at the airport network level. The models are trained using operational and weather data collected from 215 major airports across the United States. Various types of adjacency matrices were assessed for the LSTM-GCN model to identify how traffic conditions at one airport affect operations throughout the network. The experimental findings demonstrate that the LSTM model is suitable for single-airport forecasting, while the LSTM-GCN model offers enhanced accuracy when predicting traffic across the entire airport network. When compared to the traditional ARIMA model, the proposed deep-learning approaches showed lower prediction errors, demonstrating their effectiveness for both single-airport and network-level forecasting. A case study focused on Chicago O’Hare International Airport (ORD) illustrates the benefit of including both spatial and temporal factors. The proposed method provides accurate and scalable predictions, facilitating more informed decision-making in air traffic management.