Prediction of aircraft estimated time of arrival and trajectory using random forests and deep neural networks for improvement of air traffic flow

In this study, arrival times and trajectories of flights to Istanbul Ataturk International Airport which resides in Yeşilköy Terminal Maneuvering Area, were predicted using random forests and deep neural network methods. General flight information, trajectory data and meteorological aerodrome reports for arrivals to Istanbul Ataturk International Airport between April-August 2018 were obtained from different data sources and training/test data sets to be used in machine learning methods were generated. Random forests and deep neural network methods were used for the prediction of arrival times and machine learning models were obtained using data for 51.902 flights. The machine learning models decreased the prediction errors with respect to the scheduled time of departure in 64% immediately after departure and 84% after entrance to terminal maneuvering area. Similarly, deep learning method was used for trajectory prediction and machine learning model was trained with 187.509 trajectory points. The accuracy of the model was measured using the degree of fitness (R^2) and values close to 1 (0.96-0.97) were obtained for the predicted outputs which correspond to a high degree of fitness.