关键词: 经典机器学习, 深度学习, 大型机场, 离港滑行时间预测

Abstract: To improve the accuracy of taxi-out time prediction for large airports, this paper first identifies common featurevariables of taxi-out time at large airports through Pearson correlation coefficient analysis. Then, it comparespredictions of taxi-out time at Bejing Daxing International Airport and Hong Kong Chek Lap Kok InternationalAirport using classical machine leamning and deep leaming models. Experimental results show that, except for therandom forest (RF) model, the prediction results of each model are relatively close and perform well, with an av-erage prediction accuracy of 88.485%, an average prediction accuracy of 78.605% within 3 minutes, and an av-erage prediction accuracy of 93.867% within +5 minutes. The gradient boosting regression tree (GBRT) and sup-port vector regression (SVR) models outperform the others and are the best-performing models, and for the samebest-performing model, the prediction results between the two airports show litle difference. The common featurevariables of taxi-out time proposed in this paper can accurately predict taxi-out time at large airports. Classicalmachine leaming models achieve better prediction performance than deep leamning models, and the best-per-forming model demonstrates transferability.

Key words: classical machine leaming, deep leaming, large airports, taxi-out time prediction