Due to the crucial impact on flight scheduling, airline planning, and airport
operations, flight departure delay prediction has emerged as a severe and
prominent issue within the realm of smart aviation systems. Accurately
predicting flight departure delay durations constitutes a crucial aspect of
smart aviation management. Such predictive capability empowers aviation
authorities and airport regulators to implement optimized air traffic control
strategies, mitigating delays and elevating airport operational efficiency,
while enhancing the satisfaction of travelers. The methodology employed in
flight delay prediction has undergone substantial evolution in recent years,
progressing from rudimentary statistical models to more sophisticated and
intricate machine learning models. In this study, we introduce a novel machine
learning model enriched with network features and grid search-based parameter
selection for advanced predictive analytics of flight departure delays. This
model integrates air traffic network feature extraction, feature selection, and
machine learning-based prediction. Specifically, we leverage complex network
theory to extract both node-level and edge-level features from the air traffic
network. Subsequently, the XGBoost algorithm is employed for feature selection
and delay prediction, capitalizing on its flexibility and robust performance. A
case study utilizing a high-dimensional flight dataset from the U.S. Bureau of
Transportation Statistics (BTS) was conducted to assess the model’s
effectiveness. The experimental results and the visualization results
demonstrate that the proposed framework surpasses several benchmark models,
achieving an average delay prediction accuracy with a deviation of about 3.7
minutes. This framework exhibits strong potential for addressing
high-dimensional, large-scale predictive challenges in flight delay management
while maintaining superior accuracy.