In order to break through the prediction bottleneck of existing holistic modeling methods for aircraft taxiing time and improve the accuracy of ramp control-related scheduling decisions, a phased aircraft taxiing time prediction method based on phase decoupling was used to investigate the taxi process. The taxi process was divided into sub-phases with distinct dynamic characteristics (pushback and taxiway travel), and two machine learning algorithms (LightGBM and XGBoost) were adopted for independent modeling of each phase: the former realized rapid identification of key factors through an efficient gradient boosting mechanism, while the latter optimized the fitting of complex nonlinear relationships via a regularization strategy. The results show that the proposed model outperforms traditional holistic prediction in indicators such as mean absolute error (MAE) and mean squared error (MSE), and maintains high accuracy and robustness even under extreme operating conditions. It is concluded that the research results not only provide accurate predictions of aircraft taxiing time for each phase but also offers quantitative support for airport surface situational awareness and real-time optimization of full-process operations, effectively enhancing the resilience of the airport ground operation system in responding to external disturbances.