Importance and Prunability Analysis of Basic Features in Machine-Learned Aircraft Taxi-Out Time Prediction

An earlier study of the author combined the actual data of Beijing Capital International Airport (BCIA) to compare and analyze the effects of 10 machine learning methods on the flight taxi-out time prediction. The results showed that GBRT had the best prediction effect. This paper applies the GBRT method and dataset in the previous study to further discuss and analyze the importance and prunability of 16 basic features. We found that the surface co-taxiing pattern and the runway operation mode are the key factors affecting the taxi-out time of flights at BCIA, while flight attributes, air traffic flow control, and the number of flights leaving the departure apron before the target flight have little influence on the prediction results. This study can provide a scientific method for the compression of feature set, which in turn helps machine learning techniques to better integrate into flight push-back optimization.