[Purposes]This paper aims to accurately predict the flight time of aircraft during the departure phase to improve air traffic management efficiency and ensure the safety and economy of aircraft operations. [Methods]The flight characteristics of aircraft during the departure phase were deeply analyzed, and various factors influencing the flight time of departing aircraft were discussed. On this basis, characteristic waypoints were extracted, and a stacking model algorithm was established to optimize the flight time prediction for departing aircraft. The stacking model integrates neural networks, support vector machines, and linear models as base learners to enhance prediction accuracy. Meanwhile, the Bayesian ridge regression model was adopted as the meta-learner to optimize the final results. Additionally, ablation experiments were designed. Base learners were sequentially removed, and stacked sub-models containing the remaining base learners were built to compare the prediction performance of each model and deeply explore the complementary mechanisms among different models, with the feature importance of various characteristics during aircraft operations also calculated. [Findings]The stacking model outperforms single base learners in flight time prediction tasks, with an average coefficient of determination improvement of 18.8%. Meanwhile, different departure directions impose varying requirements on aircraft altitude and latitude. [Conclusions]Compared to single models, the stacking model demonstrates superior generalization capability and accuracy.