Proactive aircraft turnaround buffer optimization integrating machine learning and scenario analysis

Turnaround uncertainty is a major contributor to airline schedule unreliability and the propagation of delays across flight networks. This study proposes a learning-enhanced optimization framework for proactive aircraft turnaround buffer allocation that integrates data-driven delay predictions into planning decisions. Flight-level turnaround risk is estimated using an Extreme Gradient Boosting (XGBoost) regression model trained on historical operational data, with uncertainty captured through scenario generation based on prediction residuals. These scenarios are incorporated into a stochastic mixed-integer linear programming model that determines how limited buffer time should be allocated across flights to balance expected delay costs, buffer-related utilization costs, and environmental impacts associated with extended ground operations. The proposed framework is evaluated using real operational data from multiple airlines and aircraft types. The results show that optimal turnaround buffering is highly heterogeneous and strongly influenced by aircraft utilization patterns and exposure to delay accumulation, rather than uniform scheduling rules or airline size. Sensitivity analysis further demonstrates that buffer allocation responds coherently to economic trade-offs between delay penalties and schedule efficiency, with nonlinear transitions across buffering regimes. Overall, the study illustrates how machine-learning-derived uncertainty can be embedded into prescriptive optimization models to support proactive turnaround planning. The proposed approach provides a practical predictive–prescriptive decision-support tool for improving robustness in airline ground operations under uncertainty.