Artificial intelligence–supported weather and traffic analysis: Flight delay prediction at Istanbul Airport

Abstract

Flight delays, which directly impact operational efficiency, economic performance, and passenger satisfaction, remain a significant issue for airlines and passengers. This study aims to develop an artificial intelligence (AI)-supported framework to predict flight delays by integrating meteorological and air traffic data, focusing on Istanbul Airport (LTFM). In this study, focused on Istanbul Airport (LTFM) and conducted by integrating meteorological and air traffic data, the goal is to develop an AI-supported framework to predict flight delays. By leveraging a range of data sources, such as EUROCONTROL DDR2 flight operations, ERA5 reanalysis, and Open-Meteo forecasts, the study assembles a detailed dataset that merges environmental and operational characteristics. Various machine learning techniques—including Random Forest, XGBoost, LSTM, and Support Vector Machines—were employed to discern both linear and non-linear interactions. The models underwent evaluation using statistical performance metrics such as RMSE, MAE, R2, Accuracy, and F1-Score, in conjunction with explainability methodologies like SHAP and feature importance visualizations. The results indicate that ensemble and deep learning models surpass conventional methods in terms of predictive precision. The suggested framework not only improves delay forecasting capabilities but also delivers interpretable insights for decision-makers, thereby fostering a more robust and efficient air traffic management system.