Predicting Runway Configurations at Airports Through Soft Voting Ensemble Learning

The runway is a crucial area for airport operations, and air traffic controllers determine the current runway configuration by selecting a subset of available runways and their directions. As the number of runways at an airport increases, the complexity of choosing an appropriate runway configuration for the current weather conditions and traffic volume also increases. Establishing a runway configuration assistance decision-making system suitable for multi-runway airports is of significant practical importance. This paper utilizes actual historical data from Hartsfield-Jackson Atlanta International Airport, which has five runways. By processing meteorological data, runway configurations, and relevant operational data, the paper predicts runway configurations for twelve-time points in the next six hours, with predictions made every hour. A comparative analysis is conducted on the prediction results using three basic models: XGBoost, Random Forest, and CatBoost. Based on the results, the better model is selected and the optimal combination of weights is found for it to build the soft voter model. The research focuses on providing timely and accurate assistance decision-making recommendations for runway configurations based on meteorological and historical operational data. The aim is to enhance runway operational capacity, improve airport operational efficiency, and provide theoretical references for the development of operations at airports with multiple runways.