A Graph Neural Network Approach for Predicting Flight Delays and Inferring Delay Causes

Flight delays pose a critical problem in aviation, affecting passenger satisfaction, airline performance, and airport activities. Conventional delay forecasting models are inadequate because of the intricate interaction between weather, air traffic density, and airport-related factors. This project proposes an AIdriven flight delay forecasting system combining Graph Neural Networks (GNNs) and weather-informed modeling for improved forecasting accuracy. Flight routes and airports are modeled as a graph structure, preserving spatiotemporal relations among nearby flights. Dynamic weather conditions and past delay patterns are included in the model to enhance predictive performance. Through GraphSAGE-based feature aggregation, the system can effectively learn delay propagation patterns, allowing real-time adaptation to changing airport and weather conditions. This graph learning technique is improving the generalizability of models in various airports and making them reliable for various working conditions. This framework supports anticipatory decision-making by the airlines and airport administration, improving the efficiency of schedules and minimizing disturbance in air travel. The model can achieve test accuracy of 90.01% and exhibit excellent reliability and performance in the real world.