Predictive Modeling of Airport Flight Delays Using AI Techniques

Air transport is a critical component of global economy, facilitating business, tourism, and logistics. However, flight delays pose significant challenges, leading to economic losses and passenger dissatisfaction. Recent studies employ machine/deep learning for delay prediction but largely rely on single-source datasets such as schedules and weather, treat arrival and departure delays independently, and rarely integrate operational processes like baggage handling. This study introduces a holistic multi-source data integration framework for predictive modeling of airport flight delays using advanced Artificial Intelligence (AI) techniques. The framework combines flight schedules, Automatic Dependent Surveillance–Broadcast (ADS-B) data, weather conditions, airport-specific features, and baggage-handling information to build comprehensive predictive models for arrival and departure delays. Our primary contribution is a novel integration methodology and the introduction of the “Dissimilarity Ratio,” a baggage-derived feature that enhances prediction accuracy. For arrival delay prediction, Random Forest demonstrates superior short-term performance, achieving Mean Absolute Error (MAE) of 2.23 minutes. For departure delay prediction, XGBoost is the optimal model, with baggage-specific features improving Mean Squared Error (MSE) by up to 24.97%. These results show measurable improvements over prior studies by leveraging multi-layered operational data within a unified predictive framework and offer practical implications for improving airport efficiency and passenger satisfaction.