Tamanho da fonte: 

Despite the widespread use of machine learning (ML) methods in a variety of studies in the aviation industry in recent years, there has been limited progress in achieving explainability for the results generated by these models. Given the importance of explainable predictive models for decision support in the context of Air Traffic Management (ATM), this study explores the use of Multilayer Perceptron (MLP) and Random Forests (RF) models, along with Local Interpretable Model-agnostic Explanations (LIME), to generate local-interpretable predictions of active runway configurations and dynamic airport capacity within a 24-hour forecasting horizon for São Paulo/Guarulhos International Airport (SBGR) based on historical weather, flight demand and actual air traffic movement data. The predictive models were able to generate runway configuration forecasts with an accuracy higher than 94% and arrival/departure capacity forecasts with errors as low as 0.706. With LIME, we were able to obtain consistent explanations through the models on each prediction. An interactive predictive tool was created to output the runway configuration and capacity forecasts as well as the main contributing factors for each forecast. This research aims to improve decision-making by integrating predictive models and explanatory models into applications that provide useful information for ATM.