Departure aircraft taxi time prediction based on gradient boosting regression tree

Abstract

The rough estimation of aircraft taxi-out time causes the waste of runway slot and the low utilization efficiency. Therefore, it is urgent to propose a precise taxi time estimation method. Using the current operational data of the airport, combined with data mining to find potential factors affecting aircraft taxiing time. First, considering the large data dimensions, complexity of the algorithm and prediction inaccuracy caused by redundant learning samples, a feature selection model for optimal set of feature variables is established based on Gradient Boosted Tree (GBT). Then, the taxi-out time is predicted based on Gradient Boosting Regression Tree (GBRT). Finally, the simulation which takes Shanghai Pudong Airport as an example shows that the prediction accuracy is 76.29% within +3min, and up to 94.31% within +5min, which verifies that our method has better performance on prediction accuracy than other regression methods.