SGDAN-A Spatio-Temporal Graph Dual-Attention Neural Network for Quantified Flight Delay Prediction

Area of Science:

• Artificial Intelligence
• Transportation Science
• Data Science

Background:

• Accurate flight delay prediction is essential for passengers and airline operations.
• Estimating departure delays three hours before scheduled departure is particularly challenging yet valuable.
• Existing models often struggle with real-time conditions and complex air traffic dynamics.

Purpose of the Study:

• To develop a novel deep learning model for accurate flight departure delay estimation.
• To provide reliable delay predictions three hours prior to scheduled departure times.
• To enhance the planning capabilities for passengers and airline/airport staff.

Main Methods:

• A Spatio-temporal Graph Dual-Attention Neural Network (SGDAN) was developed.
• The air traffic network is modeled as heterogeneous graph sequences.
• Graph-level and sequence-level attention mechanisms were employed to capture inter-flight dependencies.

Main Results:

• SGDAN effectively models flight interdependencies using graph and sequence attention.
• The model generates robust node embeddings for delay time estimation.
• Experiments on a large-scale dataset demonstrate superior performance over state-of-the-art models.

Conclusions:

• SGDAN offers a significant advancement in accurate flight delay time estimation.
• The model's ability to leverage spatio-temporal information improves prediction accuracy.
• This approach provides a valuable tool for optimizing air traffic management and passenger experience.